function ConnectButton(){
console.log("Connect pushed");
document.querySelector("colab-connect-button").shadowRoot.querySelector("#connect").click()
}
setInterval(ConnectButton,6000);
In [ ]:
!nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.
In [ ]:
!pip install matplotlib==3.3.3
Collecting matplotlib==3.3.3
Downloading https://files.pythonhosted.org/packages/30/f2/10c822cb0ca5ebec58bd1892187bc3e3db64a867ac26531c6204663fc218/matplotlib-3.3.3-cp37-cp37m-manylinux1_x86_64.whl (11.6MB)
|████████████████████████████████| 11.6MB 249kB/s
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.3 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.3.3) (2.4.7)
Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.3.3) (1.3.1)
Requirement already satisfied: numpy>=1.15 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.3.3) (1.19.5)
Requirement already satisfied: pillow>=6.2.0 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.3.3) (7.1.2)
Requirement already satisfied: python-dateutil>=2.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.3.3) (2.8.1)
Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.3.3) (0.10.0)
Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.7/dist-packages (from python-dateutil>=2.1->matplotlib==3.3.3) (1.15.0)
ERROR: albumentations 0.1.12 has requirement imgaug<0.2.7,>=0.2.5, but you'll have imgaug 0.2.9 which is incompatible.
Installing collected packages: matplotlib
Found existing installation: matplotlib 3.2.2
Uninstalling matplotlib-3.2.2:
Successfully uninstalled matplotlib-3.2.2
Successfully installed matplotlib-3.3.3
In [ ]:
import sys
import os
from copy import copy
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib as mlt
import matplotlib.pyplot as plt
%matplotlib inline
import pickle
import cv2
import time
import itertools
import random
from sklearn.utils import shuffle
from sklearn.metrics import classification_report, confusion_matrix
In [ ]:
mlt.__version__
Out[ ]:
'3.3.3'
In [ ]:
import tensorflow as tf
print(tf.config.list_physical_devices())
In [ ]:
tf.config.list_physical_devices()
Out[ ]:
[PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU')]
In [ ]:
gpus= tf.config.experimental.list_physical_devices('GPU')
tf.config.experimental.set_memory_growth(gpus[0], True)
--------------------------------------------------------------------------- IndexError Traceback (most recent call last) <ipython-input-5-2e3188816bb7> in <module>() 1 gpus= tf.config.experimental.list_physical_devices('GPU') ----> 2 tf.config.experimental.set_memory_growth(gpus[0], True) IndexError: list index out of range
In [ ]:
import tensorflow.keras as keras
from tensorflow.keras import layers, models, optimizers
from tensorflow.keras.models import Sequential, Model
from tensorflow.keras.layers import Flatten, Dense, Conv2D, MaxPooling2D, Concatenate, Dot, Lambda, Input, Dropout,ZeroPadding2D, Activation, concatenate, BatchNormalization, Conv1D, GlobalAveragePooling2D
from tensorflow.keras.optimizers import Adam, RMSprop
from tensorflow.keras import optimizers
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau, CSVLogger
from tensorflow.keras import backend as K
In [ ]:
tf.__version__
Out[ ]:
'2.5.0'
In [ ]:
from google.colab import drive
drive.mount('/content/drive')
Mounted at /content/drive
In [ ]:
!unzip /content/drive/MyDrive/graduation\ project\ \(ML\)/datasets/BHSig260.zip
In [ ]:
!unzip /content/drive/MyDrive/graduation\ project\ \(ML\)/datasets/cedar55.zip
In [ ]:
!unzip /content/drive/MyDrive/graduation\ project\ \(ML\)/datasets/Dutch10.zip
In [ ]:
!unzip /content/drive/MyDrive/graduation\ project\ \(ML\)/datasets/kaggle30.zip
Data¶
In [ ]:
path = "dataset/train/"
In [ ]:
dir_list = os.listdir(path)
len(dir_list)
Out[ ]:
160
In [ ]:
def loadimgs(path, n=0, limit=False, limit_number=5):
X = []
y = []
user_dict = {}
curr_y = n
cat_y = n
i = 0
for user in os.listdir(path): #travel along every user file in the dataset folder
i = i+1
print("loading signature: " + user)
user_dict[user] = [curr_y,None]
signature_path = os.path.join(path,user)
category_images=[]
cat_y = cat_y + 1
for signature in os.listdir(signature_path): #travel along every image in each user file
image_path = os.path.join(signature_path, signature)
# image = cv2.imread(image_path, 0)
image = cv2.imread(image_path)
image = cv2.resize(image, (200, 100), interpolation=cv2.INTER_LANCZOS4)
category_images.append(image)
y.append(cat_y)
curr_y += 1
user_dict[user][1] = curr_y - 1
X.append(np.stack(category_images))
if i==limit_number and limit==True:
break
y = np.array(y)
X = np.stack(X)
return X,y,user_dict
In [ ]:
X,y,c=loadimgs(path)
loading signature: 001 loading signature: 002 loading signature: 003 loading signature: 004 loading signature: 005 loading signature: 006 loading signature: 007 loading signature: 008 loading signature: 009 loading signature: 010 loading signature: 011 loading signature: 012 loading signature: 013 loading signature: 014 loading signature: 015 loading signature: 016 loading signature: 017 loading signature: 018 loading signature: 019 loading signature: 020 loading signature: 021 loading signature: 022 loading signature: 023 loading signature: 024 loading signature: 025 loading signature: 026 loading signature: 027 loading signature: 028 loading signature: 029 loading signature: 030 loading signature: 031 loading signature: 032 loading signature: 033 loading signature: 034 loading signature: 035 loading signature: 036 loading signature: 037 loading signature: 038 loading signature: 039 loading signature: 040 loading signature: 041 loading signature: 042 loading signature: 043 loading signature: 044 loading signature: 045 loading signature: 046 loading signature: 047 loading signature: 048 loading signature: 049 loading signature: 050 loading signature: 051 loading signature: 052 loading signature: 053 loading signature: 054 loading signature: 055 loading signature: 056 loading signature: 057 loading signature: 058 loading signature: 059 loading signature: 060 loading signature: 061 loading signature: 062 loading signature: 063 loading signature: 064 loading signature: 065 loading signature: 066 loading signature: 067 loading signature: 068 loading signature: 069 loading signature: 070 loading signature: 071 loading signature: 072 loading signature: 073 loading signature: 074 loading signature: 075 loading signature: 076 loading signature: 077 loading signature: 078 loading signature: 079 loading signature: 080 loading signature: 081 loading signature: 082 loading signature: 083 loading signature: 084 loading signature: 085 loading signature: 086 loading signature: 087 loading signature: 088 loading signature: 089 loading signature: 090 loading signature: 091 loading signature: 092 loading signature: 093 loading signature: 094 loading signature: 095 loading signature: 096 loading signature: 097 loading signature: 098 loading signature: 099 loading signature: 100 loading signature: 101 loading signature: 102 loading signature: 103 loading signature: 104 loading signature: 105 loading signature: 106 loading signature: 107 loading signature: 108 loading signature: 109 loading signature: 110 loading signature: 111 loading signature: 112 loading signature: 113 loading signature: 114 loading signature: 115 loading signature: 116 loading signature: 117 loading signature: 118 loading signature: 119 loading signature: 120 loading signature: 121 loading signature: 122 loading signature: 123 loading signature: 124 loading signature: 125 loading signature: 126 loading signature: 127 loading signature: 128 loading signature: 129 loading signature: 130 loading signature: 131 loading signature: 132 loading signature: 133 loading signature: 134 loading signature: 135 loading signature: 136 loading signature: 137 loading signature: 138 loading signature: 139 loading signature: 140 loading signature: 141 loading signature: 142 loading signature: 143 loading signature: 144 loading signature: 145 loading signature: 146 loading signature: 147 loading signature: 148 loading signature: 149 loading signature: 150 loading signature: 151 loading signature: 152 loading signature: 153 loading signature: 154 loading signature: 155 loading signature: 156 loading signature: 157 loading signature: 158 loading signature: 159 loading signature: 160
In [ ]:
print(X.shape)
print(y.shape)
(160, 54, 100, 200, 3) (8640,)
In [ ]:
pickle.dump((X,y,c), open("data_load/data.pickle", "wb") )
In [ ]:
(X,y,user_dict) = pickle.load(open("data_load/data.pickle", "rb"))
X = X.astype('float32')
X = X/255
In [ ]:
print(X.shape)
print(y.shape)
(160, 54, 100, 200, 3) (8640,)
In [ ]:
i = 0
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(X[i][0],'gray')
plt.subplot(132)
plt.imshow(X[i][15],'gray')
plt.subplot(133)
plt.imshow(X[i][40],'gray')
plt.tight_layout()
plt.show()
In [ ]:
print(X.shape)
print(y.shape)
(160, 54, 100, 200, 3) (8640,)
In [ ]:
y
Out[ ]:
array([ 1, 1, 1, ..., 160, 160, 160])
In [ ]:
# user_dict
In [ ]:
X.shape
Out[ ]:
(160, 54, 100, 200, 3)
In [ ]:
X.reshape(X.shape[0]*X.shape[1],X.shape[2],X.shape[3],X.shape[4]).shape
Out[ ]:
(8640, 100, 200, 3)
In [ ]:
y.shape
Out[ ]:
(8640,)
In [ ]:
imgs = X.reshape(X.shape[0]*X.shape[1],X.shape[2],X.shape[3],X.shape[4])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(imgs[0],'gray')
plt.subplot(132)
plt.imshow(imgs[15],'gray')
plt.subplot(133)
plt.imshow(imgs[20],'gray')
plt.tight_layout()
plt.show()
Make Pairs¶
In [ ]:
print(X.reshape(X.shape[0]*X.shape[1],X.shape[2],X.shape[3],X.shape[4]).shape)
print(y.shape)
(8640, 100, 200, 3) (8640,)
In [ ]:
np.unique(y)
Out[ ]:
array([ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13,
14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,
27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39,
40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52,
53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65,
66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78,
79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91,
92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104,
105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117,
118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130,
131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143,
144, 145, 146, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156,
157, 158, 159, 160])
In [ ]:
list(range(1,6))
Out[ ]:
[1, 2, 3, 4, 5]
In [ ]:
np.where(y == 1)[0]
Out[ ]:
array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
51, 52, 53], dtype=int64)
In [ ]:
np.where(y == 160)[0]
Out[ ]:
array([8586, 8587, 8588, 8589, 8590, 8591, 8592, 8593, 8594, 8595, 8596,
8597, 8598, 8599, 8600, 8601, 8602, 8603, 8604, 8605, 8606, 8607,
8608, 8609, 8610, 8611, 8612, 8613, 8614, 8615, 8616, 8617, 8618,
8619, 8620, 8621, 8622, 8623, 8624, 8625, 8626, 8627, 8628, 8629,
8630, 8631, 8632, 8633, 8634, 8635, 8636, 8637, 8638, 8639],
dtype=int64)
In [ ]:
# make pairs
def make_pairs(x, y):
num_classes = max(y) + 1
user_indices = [np.where(y == i)[0] for i in range(1,num_classes)]
user_indices= [None] + user_indices
x = x.reshape(x.shape[0]*x.shape[1],x.shape[2],x.shape[3],x.shape[4])
pairs = []
labels = []
for idx1 in range(len(x)):
# add a matching example
x1 = x[idx1]
label1 = y[idx1]
idx2 = random.choice(user_indices[label1])
x2 = x[idx2]
pairs += [[x1, x2]]
labels += [1]
# add a not matching example
label2 = random.randint(0, num_classes-1)
while label2 == label1 or label2==0:
label2 = random.randint(0, num_classes-1)
idx2 = random.choice(user_indices[label2])
x2 = x[idx2]
pairs += [[x1, x2]]
labels += [0]
return np.array(pairs), np.array(labels)
In [ ]:
pairs_train, labels_train = make_pairs(X, y)
In [ ]:
print(pairs_train.shape)
print(labels_train.shape)
(17280, 2, 100, 200, 3) (17280,)
In [ ]:
pickle.dump((pairs_train,labels_train), open("data/train.pickle", "wb") )
In [ ]:
(pairs_train, labels_train) = pickle.load(open("data/train.pickle", "rb"))
In [ ]:
print(pairs_train.shape)
print(labels_train.shape)
(17280, 2, 100, 200, 3) (17280,)
In [ ]:
i = 267
print(labels_train[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs_train[i][0],'gray')
plt.subplot(132)
plt.imshow(pairs_train[i][1],'gray')
plt.tight_layout()
plt.show()
0
Data 2¶
In [ ]:
path = "dataset/train/"
In [ ]:
dir_list = os.listdir(path)
len(dir_list)
Out[ ]:
160
In [ ]:
data = []
for directory in dir_list:
images = os.listdir(path+directory)
images = [path+directory+'/'+x for x in images]
data.append(images)
In [ ]:
data = []
for directory in dir_list:
images = os.listdir(path+directory)
images = [path+directory+'/'+x for x in images]
data.append(images[30:])
In [ ]:
len(data)
Out[ ]:
160
In [ ]:
len(data[0])
Out[ ]:
24
In [ ]:
data[6][:]
Out[ ]:
['dataset/train/007/H-S-7-G-01.tif', 'dataset/train/007/H-S-7-G-02.tif', 'dataset/train/007/H-S-7-G-03.tif', 'dataset/train/007/H-S-7-G-04.tif', 'dataset/train/007/H-S-7-G-05.tif', 'dataset/train/007/H-S-7-G-06.tif', 'dataset/train/007/H-S-7-G-07.tif', 'dataset/train/007/H-S-7-G-08.tif', 'dataset/train/007/H-S-7-G-09.tif', 'dataset/train/007/H-S-7-G-10.tif', 'dataset/train/007/H-S-7-G-11.tif', 'dataset/train/007/H-S-7-G-12.tif', 'dataset/train/007/H-S-7-G-13.tif', 'dataset/train/007/H-S-7-G-14.tif', 'dataset/train/007/H-S-7-G-15.tif', 'dataset/train/007/H-S-7-G-16.tif', 'dataset/train/007/H-S-7-G-17.tif', 'dataset/train/007/H-S-7-G-18.tif', 'dataset/train/007/H-S-7-G-19.tif', 'dataset/train/007/H-S-7-G-20.tif', 'dataset/train/007/H-S-7-G-21.tif', 'dataset/train/007/H-S-7-G-22.tif', 'dataset/train/007/H-S-7-G-23.tif', 'dataset/train/007/H-S-7-G-24.tif']
In [ ]:
img_h, img_w, img_ch = 100, 200 , 3
In [ ]:
def visualize_sample_signature(choose=False,idx=0):
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize = (15, 10))
if choose:
k = idx
else:
k = np.random.randint(len(data))
print(k)
img_names = random.sample(data[k], 3)
print(img_names)
img1 = cv2.imread(img_names[0])
img2 = cv2.imread(img_names[1])
img3 = cv2.imread(img_names[2])
img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img3 = cv2.resize(img3, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
ax1.imshow(img1, cmap = 'gray')
ax2.imshow(img2, cmap = 'gray')
ax3.imshow(img3, cmap = 'gray')
ax1.set_title(0)
ax1.set_xlabel(0)
# ax1.axis('off')
ax2.set_title(1)
ax2.set_xlabel(1)
# ax2.axis('off')
ax3.set_title(2)
ax3.set_xlabel(2)
# ax3.axis('off')
plt.tight_layout()
plt.show()
In [ ]:
visualize_sample_signature(choose=True,idx=55)
55 ['dataset/train/056/H-S-56-G-13.tif', 'dataset/train/056/H-S-56-G-22.tif', 'dataset/train/056/H-S-56-G-08.tif']
In [ ]:
# make pairs
def generate_batch(x):
pairs = []
labels = []
for label1 in range(len(data)):
for idx1 in range(len(data[label1])):
# add a matching example
x1 = data[label1][idx1]
idx2 = random.randint(0,len(data[label1])-1)
x2 = data[label1][idx2]
pairs += [[x1, x2]]
labels += [1]
# add a not matching example
label2 = random.randint(0, len(data)-1)
while label2 == label1:
label2 = random.randint(0, len(data)-1)
idx2 = random.randint(0,len(data[label2])-1)
x2 = data[label2][idx2]
pairs += [[x1, x2]]
labels += [0]
# pairs, labels = shuffle(pairs, labels)
return np.array(pairs), np.array(labels)
In [ ]:
pairs, labels = generate_batch(data)
In [ ]:
print(pairs.shape)
print(labels.shape)
(17280, 2) (17280,)
In [ ]:
i = 2
print(labels[i])
print(pairs[i])
print(labels[i+1])
print(pairs[i+1])
1 ['dataset/train/001/H-S-1-F-02.tif' 'dataset/train/001/H-S-1-F-22.tif'] 0 ['dataset/train/001/H-S-1-F-02.tif' 'dataset/train/056/H-S-56-F-12.tif']
In [ ]:
def generate_batch(data, batch_size = 32):
while True:
all_pairs = []
labels = []
for label1 in range(len(data)):
for idx1 in range(len(data[label1])):
# add a matching example
x1 = data[label1][idx1]
idx2 = random.randint(0,len(data[label1])-1)
x2 = data[label1][idx2]
all_pairs += [[x1, x2]]
labels += [1]
# add a not matching example
label2 = random.randint(0, len(data)-1)
while label2 == label1:
label2 = random.randint(0, len(data)-1)
idx2 = random.randint(0,len(data[label2])-1)
x2 = data[label2][idx2]
all_pairs += [[x1, x2]]
labels += [0]
# all_pairs, labels = shuffle(all_pairs, labels)
k = 0
pairs=[]
targets=[]
for ix, pair in enumerate(all_pairs):
img1 = cv2.imread(pair[0])
img2 = cv2.imread(pair[1])
img1 = cv2.resize(img1, (img_w, img_h))
img2 = cv2.resize(img2, (img_w, img_h))
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
pairs.append([img1, img2])
targets.append(labels[ix])
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
In [ ]:
def generate_batch(data, batch_size = 32):
while True:
all_pairs = []
labels = []
for label1 in range(len(data)):
for idx1 in range(len(data[label1])):
# add a matching example
x1 = data[label1][idx1]
idx2 = random.randint(0,len(data[label1])-1)
x2 = data[label1][idx2]
all_pairs += [[x1, x2]]
labels += [1]
# add a not matching example
label2 = random.randint(0, len(data)-1)
while label2 == label1:
label2 = random.randint(0, len(data)-1)
idx2 = random.randint(0,len(data[label2])-1)
x2 = data[label2][idx2]
all_pairs += [[x1, x2]]
labels += [0]
# all_pairs, labels = shuffle(all_pairs, labels)
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
for ix, pair in enumerate(all_pairs):
img1 = cv2.imread(pair[0])
img2 = cv2.imread(pair[1])
img1 = cv2.resize(img1, (img_w, img_h))
img2 = cv2.resize(img2, (img_w, img_h))
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
pairs[0][k, :, :, :] = img1
pairs[1][k, :, :, :] = img2
targets[k] = labels[ix]
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
visualize generate_batch¶
In [ ]:
ff=generate_batch(data, batch_size = 32)
pairs, targets = next(ff)
pairs, targets = next(ff)
# pairs, targets = next(ff)
# pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
print(pairs.shape)
print(targets.shape)
(2, 32, 100, 200, 3) (32,)
In [ ]:
print(pairs[0].shape)
(32, 100, 200, 3)
In [ ]:
print(targets)
[1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0.]
In [ ]:
i = 15
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
0.0
In [ ]:
i = 5
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[i][0],'gray')
plt.subplot(132)
plt.imshow(pairs[i][1],'gray')
plt.tight_layout()
plt.show()
0.0
--------------------------------------------------------------------------- IndexError Traceback (most recent call last) <ipython-input-24-99a41ad785ce> in <module> 5 plt.figure(figsize = (15, 10)) 6 plt.subplot(131) ----> 7 plt.imshow(pairs[i][0],'gray') 8 plt.subplot(132) 9 plt.imshow(pairs[i][1],'gray') IndexError: index 5 is out of bounds for axis 0 with size 2
Data 3¶
In [ ]:
# !unzip /content/drive/MyDrive/graduation\ project\ \(ML\)/datasets/BHSig260.zip
In [ ]:
# path = "dataset/train/"
In [ ]:
path1 = "/content/BHSig260/Hindi/"
path2 = "/content/BHSig260/Bengali/"
In [ ]:
img_h, img_w, img_ch = 150, 300, 1
In [ ]:
# dir_list = os.listdir(path)
# len(dir_list)
In [ ]:
dir_list1 = next(os.walk(path1))[1]
dir_list1.sort()
len(dir_list1)
Out[ ]:
160
In [ ]:
dir_list2 = next(os.walk(path2))[1]
dir_list2.sort()
len(dir_list2)
Out[ ]:
100
In [ ]:
dir_list1 = shuffle(dir_list1)
dir_list2 = shuffle(dir_list2)
In [ ]:
dir_list1[:20], dir_list2[:20]
Out[ ]:
(['100', '012', '149', '087', '035', '107', '054', '151', '061', '043', '111', '007', '156', '145', '025', '130', '154', '157', '138', '070'], ['058', '017', '088', '069', '090', '085', '001', '023', '020', '100', '008', '014', '081', '096', '005', '031', '009', '068', '027', '029'])
In [ ]:
orig_groups, forg_groups = [], []
In [ ]:
for i,directory in enumerate(dir_list1):
if i==20:
break
images = os.listdir(path1+directory)
images.sort()
images = [path1+directory+'/'+x for x in images]
forg_groups.append(images[:30])
orig_groups.append(images[30:])
In [ ]:
for i,directory in enumerate(dir_list2):
if i==20:
break
images = os.listdir(path2+directory)
images.sort()
images = [path2+directory+'/'+x for x in images]
forg_groups.append(images[:30])
orig_groups.append(images[30:])
In [ ]:
print(len(orig_groups))
print(len(forg_groups))
print(len(orig_groups[0]))
print(len(forg_groups[0]))
40 40 24 30
In [ ]:
orig_groups, forg_groups = shuffle(orig_groups, forg_groups)
In [ ]:
orig_groups[10], forg_groups[10]
Out[ ]:
(['/content/BHSig260/Bengali/036/B-S-36-G-01.tif', '/content/BHSig260/Bengali/036/B-S-36-G-02.tif', '/content/BHSig260/Bengali/036/B-S-36-G-03.tif', '/content/BHSig260/Bengali/036/B-S-36-G-04.tif', '/content/BHSig260/Bengali/036/B-S-36-G-05.tif', '/content/BHSig260/Bengali/036/B-S-36-G-06.tif', '/content/BHSig260/Bengali/036/B-S-36-G-07.tif', '/content/BHSig260/Bengali/036/B-S-36-G-08.tif', '/content/BHSig260/Bengali/036/B-S-36-G-09.tif', '/content/BHSig260/Bengali/036/B-S-36-G-10.tif', '/content/BHSig260/Bengali/036/B-S-36-G-11.tif', '/content/BHSig260/Bengali/036/B-S-36-G-12.tif', '/content/BHSig260/Bengali/036/B-S-36-G-13.tif', '/content/BHSig260/Bengali/036/B-S-36-G-14.tif', '/content/BHSig260/Bengali/036/B-S-36-G-15.tif', '/content/BHSig260/Bengali/036/B-S-36-G-16.tif', '/content/BHSig260/Bengali/036/B-S-36-G-17.tif', '/content/BHSig260/Bengali/036/B-S-36-G-18.tif', '/content/BHSig260/Bengali/036/B-S-36-G-19.tif', '/content/BHSig260/Bengali/036/B-S-36-G-20.tif', '/content/BHSig260/Bengali/036/B-S-36-G-21.tif', '/content/BHSig260/Bengali/036/B-S-36-G-22.tif', '/content/BHSig260/Bengali/036/B-S-36-G-23.tif', '/content/BHSig260/Bengali/036/B-S-36-G-24.tif'], ['/content/BHSig260/Bengali/036/B-S-36-F-01.tif', '/content/BHSig260/Bengali/036/B-S-36-F-02.tif', '/content/BHSig260/Bengali/036/B-S-36-F-03.tif', '/content/BHSig260/Bengali/036/B-S-36-F-04.tif', '/content/BHSig260/Bengali/036/B-S-36-F-05.tif', '/content/BHSig260/Bengali/036/B-S-36-F-06.tif', '/content/BHSig260/Bengali/036/B-S-36-F-07.tif', '/content/BHSig260/Bengali/036/B-S-36-F-08.tif', '/content/BHSig260/Bengali/036/B-S-36-F-09.tif', '/content/BHSig260/Bengali/036/B-S-36-F-10.tif', '/content/BHSig260/Bengali/036/B-S-36-F-11.tif', '/content/BHSig260/Bengali/036/B-S-36-F-12.tif', '/content/BHSig260/Bengali/036/B-S-36-F-13.tif', '/content/BHSig260/Bengali/036/B-S-36-F-14.tif', '/content/BHSig260/Bengali/036/B-S-36-F-15.tif', '/content/BHSig260/Bengali/036/B-S-36-F-16.tif', '/content/BHSig260/Bengali/036/B-S-36-F-17.tif', '/content/BHSig260/Bengali/036/B-S-36-F-18.tif', '/content/BHSig260/Bengali/036/B-S-36-F-19.tif', '/content/BHSig260/Bengali/036/B-S-36-F-20.tif', '/content/BHSig260/Bengali/036/B-S-36-F-21.tif', '/content/BHSig260/Bengali/036/B-S-36-F-22.tif', '/content/BHSig260/Bengali/036/B-S-36-F-23.tif', '/content/BHSig260/Bengali/036/B-S-36-F-24.tif', '/content/BHSig260/Bengali/036/B-S-36-F-25.tif', '/content/BHSig260/Bengali/036/B-S-36-F-26.tif', '/content/BHSig260/Bengali/036/B-S-36-F-27.tif', '/content/BHSig260/Bengali/036/B-S-36-F-28.tif', '/content/BHSig260/Bengali/036/B-S-36-F-29.tif', '/content/BHSig260/Bengali/036/B-S-36-F-30.tif'])
In [ ]:
orig_lengths = [len(x) for x in orig_groups]
forg_lengths = [len(x) for x in forg_groups]
print(orig_lengths)
print(forg_lengths)
[24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24] [30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30]
In [ ]:
# orig_train, orig_val, orig_test = orig_groups[:120], orig_groups[120:140], orig_groups[140:]
# forg_train, forg_val, forg_test = forg_groups[:120], forg_groups[120:140], forg_groups[140:]
In [ ]:
# orig_val, orig_test, orig_train = orig_groups[0:20], orig_groups[20:40], orig_groups[40:]
# forg_val, forg_test, forg_train = forg_groups[0:20], forg_groups[20:40], forg_groups[40:]
In [ ]:
# orig_train, orig_val = orig_groups[:220], orig_groups[220:]
# forg_train, forg_val = forg_groups[:220], forg_groups[220:]
In [ ]:
orig_train, orig_val = orig_groups[25:85], orig_groups[240:]
forg_train, forg_val = forg_groups[25:85], forg_groups[240:]
In [ ]:
len(orig_train), len(orig_val)
Out[ ]:
(60, 20)
In [ ]:
# forg_groups[0]
In [ ]:
orig_train[0], orig_val[0]
Out[ ]:
(['/content/BHSig260/Hindi/100/H-S-100-G-01.tif', '/content/BHSig260/Hindi/100/H-S-100-G-02.tif', '/content/BHSig260/Hindi/100/H-S-100-G-03.tif', '/content/BHSig260/Hindi/100/H-S-100-G-04.tif', '/content/BHSig260/Hindi/100/H-S-100-G-05.tif', '/content/BHSig260/Hindi/100/H-S-100-G-06.tif', '/content/BHSig260/Hindi/100/H-S-100-G-07.tif', '/content/BHSig260/Hindi/100/H-S-100-G-08.tif', '/content/BHSig260/Hindi/100/H-S-100-G-09.tif', '/content/BHSig260/Hindi/100/H-S-100-G-10.tif', '/content/BHSig260/Hindi/100/H-S-100-G-11.tif', '/content/BHSig260/Hindi/100/H-S-100-G-12.tif', '/content/BHSig260/Hindi/100/H-S-100-G-13.tif', '/content/BHSig260/Hindi/100/H-S-100-G-14.tif', '/content/BHSig260/Hindi/100/H-S-100-G-15.tif', '/content/BHSig260/Hindi/100/H-S-100-G-16.tif', '/content/BHSig260/Hindi/100/H-S-100-G-17.tif', '/content/BHSig260/Hindi/100/H-S-100-G-18.tif', '/content/BHSig260/Hindi/100/H-S-100-G-19.tif', '/content/BHSig260/Hindi/100/H-S-100-G-20.tif', '/content/BHSig260/Hindi/100/H-S-100-G-21.tif', '/content/BHSig260/Hindi/100/H-S-100-G-22.tif', '/content/BHSig260/Hindi/100/H-S-100-G-23.tif', '/content/BHSig260/Hindi/100/H-S-100-G-24.tif'], ['/content/BHSig260/Bengali/082/B-S-82-G-01.tif', '/content/BHSig260/Bengali/082/B-S-82-G-02.tif', '/content/BHSig260/Bengali/082/B-S-82-G-03.tif', '/content/BHSig260/Bengali/082/B-S-82-G-04.tif', '/content/BHSig260/Bengali/082/B-S-82-G-05.tif', '/content/BHSig260/Bengali/082/B-S-82-G-06.tif', '/content/BHSig260/Bengali/082/B-S-82-G-07.tif', '/content/BHSig260/Bengali/082/B-S-82-G-08.tif', '/content/BHSig260/Bengali/082/B-S-82-G-09.tif', '/content/BHSig260/Bengali/082/B-S-82-G-10.tif', '/content/BHSig260/Bengali/082/B-S-82-G-11.tif', '/content/BHSig260/Bengali/082/B-S-82-G-12.tif', '/content/BHSig260/Bengali/082/B-S-82-G-13.tif', '/content/BHSig260/Bengali/082/B-S-82-G-14.tif', '/content/BHSig260/Bengali/082/B-S-82-G-15.tif', '/content/BHSig260/Bengali/082/B-S-82-G-16.tif', '/content/BHSig260/Bengali/082/B-S-82-G-17.tif', '/content/BHSig260/Bengali/082/B-S-82-G-18.tif', '/content/BHSig260/Bengali/082/B-S-82-G-19.tif', '/content/BHSig260/Bengali/082/B-S-82-G-20.tif', '/content/BHSig260/Bengali/082/B-S-82-G-21.tif', '/content/BHSig260/Bengali/082/B-S-82-G-22.tif', '/content/BHSig260/Bengali/082/B-S-82-G-23.tif', '/content/BHSig260/Bengali/082/B-S-82-G-24.tif'])
In [ ]:
# # Delete unnecessary variables
# del orig_groups, forg_groups
In [ ]:
def visualize_sample_signature(choose=False,idx=0):
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize = (14, 7))
if choose:
k = idx
else:
k = np.random.randint(len(orig_groups))
print(k)
orig_img_names = random.sample(orig_groups[k], 2)
forg_img_name = random.sample(forg_groups[k], 1)
print(orig_img_names+forg_img_name)
orig_img1 = cv2.imread(orig_img_names[0])
orig_img2 = cv2.imread(orig_img_names[1])
forg_img = plt.imread(forg_img_name[0])
orig_img1 = cv2.resize(orig_img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
orig_img2 = cv2.resize(orig_img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
forg_img = cv2.resize(forg_img, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
ax1.imshow(orig_img1, cmap = 'gray')
ax2.imshow(orig_img2, cmap = 'gray')
ax3.imshow(forg_img, cmap = 'gray')
ax1.set_title('Genuine Copy')
ax1.set_xlabel(1)
# ax1.axis('off')
ax2.set_title('Genuine Copy')
ax2.set_xlabel(2)
# ax2.axis('off')
ax3.set_title('Forged Copy')
ax3.set_xlabel(3)
# ax3.axis('off')
plt.tight_layout()
plt.show()
In [ ]:
visualize_sample_signature(choose=True,idx=0)
0 ['/content/BHSig260/Bengali/016/B-S-16-G-14.tif', '/content/BHSig260/Bengali/016/B-S-16-G-12.tif', '/content/BHSig260/Bengali/016/B-S-16-F-04.tif']
In [ ]:
num_samples = 0
pairss = []
In [ ]:
num_samples
Out[ ]:
0
In [ ]:
kernel = np.ones((9,9),np.uint8) # default
def preprocessor_img(path, image_shape):
image = cv2.imread(path,0)
blured = cv2.GaussianBlur(image, (9,9), 0)
threshold, binary = cv2.threshold(blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel, iterations=20)
contours, hierarchies = cv2.findContours(closing, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
the_biggest_contour_by_area = max(contours, key=cv2.contourArea)
x,y,w,h = cv2.boundingRect(the_biggest_contour_by_area)
cropped = image[y:y+h, x:x+w]
resized = cv2.resize(cropped, image_shape, interpolation=cv2.INTER_LANCZOS4)
resized_blured = cv2.GaussianBlur(resized, (9,9), 0)
threshold, resized_binary = cv2.threshold(resized_blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
return resized_binary
In [ ]:
def generate_batch(orig_data, forg_data, batch_size = 32):
# global num_samples, pairss
while True:
orig_pairs = []
forg_pairs = []
gen_gen_labels = []
gen_for_labels = []
all_pairs = []
all_labels = []
# Here we create pairs of Genuine-Genuine image names and Genuine-Forged image names
# For every person we have 24 genuine signatures, hence we have
# 24 choose 2 = 276 Genuine-Genuine image pairs for one person.
# To make Genuine-Forged pairs, we pair every Genuine signature of a person
# with 12 randomly sampled Forged signatures of the same person.
# Thus we make 24 * 12 = 288 Genuine-Forged image pairs for one person.
# In all we have 120 person's data in the training data.
# Total no. of Genuine-Genuine pairs = 160 * 276 = 44160
# Total number of Genuine-Forged pairs = 160 * 288 = 46080
# Total no. of data points = 44160 + 46080 = 90240
for orig, forg in zip(orig_data, forg_data):
orig_pairs.extend(list(itertools.combinations(orig, 2)))
for i in range(len(forg)):
forg_pairs.extend(list(itertools.product(orig[i:i+1], random.sample(forg, 12))))
# Label for Genuine-Genuine pairs is 1
# Label for Genuine-Forged pairs is 0
gen_gen_labels = [1]*len(orig_pairs)
gen_for_labels = [0]*len(forg_pairs)
# Concatenate all the pairs together along with their labels and shuffle them
all_pairs = orig_pairs + forg_pairs
all_labels = gen_gen_labels + gen_for_labels
del orig_pairs, forg_pairs, gen_gen_labels, gen_for_labels
all_pairs, all_labels = shuffle(all_pairs, all_labels)
# print(len(all_pairs))
# num_samples = len(all_pairs)
# pairss = all_pairs
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
for ix, pair in enumerate(all_pairs):
# img1 = preprocessor_img(pair[0], (img_w, img_h))
# img2 = preprocessor_img(pair[1], (img_w, img_h))
img1 = cv2.imread(pair[0],0)
img2 = cv2.imread(pair[1],0)
img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
img1 = np.atleast_3d(img1)
img2 = np.atleast_3d(img2)
# img2 = img2[...,np.newaxis]
pairs[0][k, :, :, :] = img1
pairs[1][k, :, :, :] = img2
targets[k] = all_labels[ix]
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
visualize generate_batch¶
In [ ]:
ff = generate_batch(orig_val, forg_val, batch_size = 32)
pairs, targets = next(ff)
# pairs, targets = next(ff)
# pairs, targets = next(ff)
# pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
# print(num_samples)
In [ ]:
# del pairs,targets
In [ ]:
print(pairs.shape)
print(targets.shape)
(2, 32, 150, 300, 1) (32,)
In [ ]:
print(pairs[0].shape)
(32, 150, 300, 1)
In [ ]:
print(targets)
[1. 1. 0. 1. 1. 0. 1. 0. 1. 0. 0. 1. 0. 1. 1. 1. 1. 1. 1. 1. 1. 0. 1. 0. 1. 0. 0. 1. 0. 0. 1. 0.]
In [ ]:
i = 1
In [ ]:
# pairss[i]
In [ ]:
# pairss[4%32*1]
In [ ]:
i = 2
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
0.0
In [ ]:
In [ ]:
Data 4¶
In [ ]:
img_h, img_w, img_ch = 150, 300, 1
image_shape = (img_h, img_w, img_ch)
image_shape
Out[ ]:
(150, 300, 1)
In [ ]:
path = "/content/cedar/"
In [ ]:
dir_list = next(os.walk(path))[1]
dir_list.sort(key = int)
len(dir_list)
Out[ ]:
55
In [ ]:
dir_list = shuffle(dir_list)
In [ ]:
dir_list[:30]
Out[ ]:
['9', '35', '20', '36', '41', '23', '3', '42', '21', '43', '44', '32', '39', '34', '24', '5', '47', '16', '30', '7', '54', '13', '49', '45', '28', '6', '50', '31', '40', '38']
In [ ]:
orig_groups, forg_groups = [], []
In [ ]:
for i,directory in enumerate(dir_list):
if i==30:
break
images = os.listdir(path+directory)
images.sort()
images = [path+directory+'/'+x for x in images]
forg_groups.append(images[:24])
orig_groups.append(images[24:])
In [ ]:
print(len(orig_groups))
print(len(forg_groups))
print(len(orig_groups[0]))
print(len(forg_groups[0]))
70 70 24 30
In [ ]:
orig_lengths = [len(x) for x in orig_groups]
forg_lengths = [len(x) for x in forg_groups]
print(orig_lengths)
print(forg_lengths)
[24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24] [30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24]
In [ ]:
# orig_groups[20], forg_groups[20]
In [ ]:
orig_groups, forg_groups = shuffle(orig_groups, forg_groups)
In [ ]:
orig_lengths = [len(x) for x in orig_groups]
forg_lengths = [len(x) for x in forg_groups]
print(orig_lengths)
print(forg_lengths)
[24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24] [24, 30, 30, 30, 24, 30, 24, 24, 30, 30, 24, 30, 30, 24, 30, 24, 30, 30, 30, 30, 30, 30, 30, 24, 24, 30, 24, 30, 30, 30]
In [ ]:
# forg_groups[0], orig_groups[0]
In [ ]:
orig_train, orig_val = orig_groups[:45], orig_groups[45:]
forg_train, forg_val = forg_groups[:45], forg_groups[45:]
In [ ]:
# orig_train, orig_val = orig_groups[:9], orig_groups[9:]
# forg_train, forg_val = forg_groups[:9], forg_groups[9:]
In [ ]:
orig_train, orig_val = orig_groups[:25], orig_groups[25:]
forg_train, forg_val = forg_groups[:25], forg_groups[25:]
In [ ]:
len(orig_train), len(orig_val), len(forg_train), len(forg_val)
Out[ ]:
(25, 5, 25, 5)
In [ ]:
# orig_train[0], forg_train[0]
In [ ]:
# # Delete unnecessary variables
# del orig_groups, forg_groups
In [ ]:
def visualize_sample_signature(choose=False,idx=0):
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize = (14, 7))
if choose:
k = idx
else:
k = np.random.randint(len(orig_groups))
print(k)
orig_img_names = random.sample(orig_groups[k], 2)
forg_img_name = random.sample(forg_groups[k], 1)
print(orig_img_names+forg_img_name)
orig_img1 = cv2.imread(orig_img_names[0])
orig_img2 = cv2.imread(orig_img_names[1])
forg_img = plt.imread(forg_img_name[0])
orig_img1 = cv2.resize(orig_img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
orig_img2 = cv2.resize(orig_img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
forg_img = cv2.resize(forg_img, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
ax1.imshow(orig_img1, cmap = 'gray')
ax2.imshow(orig_img2, cmap = 'gray')
ax3.imshow(forg_img, cmap = 'gray')
ax1.set_title('Genuine Copy')
ax1.set_xlabel(1)
# ax1.axis('off')
ax2.set_title('Genuine Copy')
ax2.set_xlabel(2)
# ax2.axis('off')
ax3.set_title('Forged Copy')
ax3.set_xlabel(3)
# ax3.axis('off')
plt.tight_layout()
plt.show()
In [ ]:
visualize_sample_signature(choose=True,idx=25)
25 ['/content/BHSig260/Hindi/146/H-S-146-G-17.tif', '/content/BHSig260/Hindi/146/H-S-146-G-10.tif', '/content/BHSig260/Hindi/146/H-S-146-F-04.tif']
In [ ]:
kernel = np.ones((9,9),np.uint8) # default
def preprocessor_img(path, image_shape):
image = cv2.imread(path,0)
blured = cv2.GaussianBlur(image, (9,9), 0)
threshold, binary = cv2.threshold(blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel, iterations=20)
contours, hierarchies = cv2.findContours(closing, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
the_biggest_contour_by_area = max(contours, key=cv2.contourArea)
x,y,w,h = cv2.boundingRect(the_biggest_contour_by_area)
cropped = image[y:y+h, x:x+w]
resized = cv2.resize(cropped, image_shape, interpolation=cv2.INTER_LANCZOS4)
resized_blured = cv2.GaussianBlur(resized, (9,9), 0)
threshold, resized_binary = cv2.threshold(resized_blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
return resized_binary
In [ ]:
# num_samples=0
In [ ]:
def generate_batch(orig_data, forg_data, batch_size = 32):
# global num_samples, pairss
while True:
orig_pairs = []
forg_pairs = []
gen_gen_labels = []
gen_for_labels = []
all_pairs = []
all_labels = []
# Here we create pairs of Genuine-Genuine image names and Genuine-Forged image names
# For every person we have 24 genuine signatures, hence we have
# 24 choose 2 = 276 Genuine-Genuine image pairs for one person.
# To make Genuine-Forged pairs, we pair every Genuine signature of a person
# with 12 randomly sampled Forged signatures of the same person.
# Thus we make 24 * 12 = 288 Genuine-Forged image pairs for one person.
# In all we have 120 person's data in the training data.
# Total no. of Genuine-Genuine pairs = 160 * 276 = 44160
# Total number of Genuine-Forged pairs = 160 * 288 = 46080
# Total no. of data points = 44160 + 46080 = 90240
for orig, forg in zip(orig_data, forg_data):
orig_pairs.extend(list(itertools.combinations(orig, 2)))
for i in range(len(forg)):
forg_pairs.extend(list(itertools.product(orig[i:i+1], random.sample(forg, 12))))
# Label for Genuine-Genuine pairs is 1
# Label for Genuine-Forged pairs is 0
gen_gen_labels = [1]*len(orig_pairs)
gen_for_labels = [0]*len(forg_pairs)
# Concatenate all the pairs together along with their labels and shuffle them
all_pairs = orig_pairs + forg_pairs
all_labels = gen_gen_labels + gen_for_labels
del orig_pairs, forg_pairs, gen_gen_labels, gen_for_labels
all_pairs, all_labels = shuffle(all_pairs, all_labels)
# print(len(all_pairs))
# num_samples = len(all_pairs)
# pairss = all_pairs
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
for ix, pair in enumerate(all_pairs):
# img1 = preprocessor_img(pair[0], (img_w, img_h))
# img2 = preprocessor_img(pair[1], (img_w, img_h))
img1 = cv2.imread(pair[0],0)
img2 = cv2.imread(pair[1],0)
img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
img1 = np.atleast_3d(img1)
img2 = np.atleast_3d(img2)
# img2 = img2[...,np.newaxis]
pairs[0][k, :, :, :] = img1
pairs[1][k, :, :, :] = img2
targets[k] = all_labels[ix]
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
visualize generate_batch¶
In [ ]:
ff = generate_batch(orig_train, forg_train, batch_size = 32)
pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
# num_samples
In [ ]:
print(pairs.shape)
print(targets.shape)
(2, 32, 150, 300, 1) (32,)
In [ ]:
print(pairs[0].shape)
(32, 150, 300, 1)
In [ ]:
print(targets)
[1. 0. 0. 0. 1. 1. 1. 1. 0. 1. 1. 0. 1. 1. 0. 0. 1. 1. 0. 0. 1. 0. 1. 0. 0. 0. 1. 0. 1. 1. 1. 1.]
In [ ]:
i = 4
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
--------------------------------------------------------------------------- NameError Traceback (most recent call last) <ipython-input-1-de027d524c50> in <module>() 1 i = 4 2 ----> 3 print(targets[i]) 4 5 plt.figure(figsize = (15, 10)) NameError: name 'targets' is not defined
In [ ]:
In [ ]:
In [ ]:
In [ ]:
Data 5 -- Test dutch¶
In [ ]:
img_h, img_w, img_ch = 150, 300, 1
image_shape = (img_h, img_w, img_ch)
image_shape
Out[ ]:
(150, 300, 1)
In [ ]:
path = "/content/dutch/"
In [ ]:
dir_list = next(os.walk(path))[1]
dir_list.sort(key = int)
len(dir_list)
Out[ ]:
10
In [ ]:
orig_groups, forg_groups = [], []
In [ ]:
for i,directory in enumerate(dir_list):
# if i==10:
# break
images = os.listdir(path+directory)
images.sort()
images = [path+directory+'/'+x for x in images]
forg_groups.append(images[:24])
orig_groups.append(images[24:])
In [ ]:
print(len(forg_groups))
print(len(orig_groups))
print(len(forg_groups[0]))
print(len(orig_groups[0]))
10 10 24 12
In [ ]:
forg_lengths = [len(x) for x in forg_groups]
orig_lengths = [len(x) for x in orig_groups]
print(forg_lengths)
print(orig_lengths)
[24, 24, 24, 24, 24, 24, 24, 24, 24, 24] [12, 12, 12, 12, 12, 12, 12, 12, 12, 12]
In [ ]:
# orig_groups[30], forg_groups[30]
In [ ]:
def visualize_sample_signature(choose=False,idx=0):
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize = (14, 7))
if choose:
k = idx
else:
k = np.random.randint(len(orig_groups))
print(k)
orig_img_names = random.sample(orig_groups[k], 2)
forg_img_name = random.sample(forg_groups[k], 1)
print(orig_img_names+forg_img_name)
orig_img1 = cv2.imread(orig_img_names[0],0)
orig_img2 = cv2.imread(orig_img_names[1],0)
forg_img = cv2.imread(forg_img_name[0],0)
orig_img1 = cv2.resize(orig_img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
orig_img2 = cv2.resize(orig_img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
forg_img = cv2.resize(forg_img, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
ax1.imshow(orig_img1, cmap = 'gray')
ax2.imshow(orig_img2, cmap = 'gray')
ax3.imshow(forg_img, cmap = 'gray')
ax1.set_title('Genuine Copy')
ax1.set_xlabel(1)
# ax1.axis('off')
ax2.set_title('Genuine Copy')
ax2.set_xlabel(2)
# ax2.axis('off')
ax3.set_title('Forged Copy')
ax3.set_xlabel(3)
# ax3.axis('off')
plt.tight_layout()
plt.show()
In [ ]:
visualize_sample_signature(choose=True,idx=7)
7 ['/content/dutch/08/14_024.PNG', '/content/dutch/08/19_024.PNG', '/content/dutch/08/04_0120024.PNG']
In [ ]:
kernel = np.ones((9,9),np.uint8) # default
def preprocessor_img(path, image_shape):
image = cv2.imread(path,0)
blured = cv2.GaussianBlur(image, (9,9), 0)
threshold, binary = cv2.threshold(blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel, iterations=30)
contours, hierarchies = cv2.findContours(closing, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
the_biggest_contour_by_area = max(contours, key=cv2.contourArea)
x,y,w,h = cv2.boundingRect(the_biggest_contour_by_area)
cropped = image[y:y+h, x:x+w]
resized = cv2.resize(cropped, image_shape, interpolation=cv2.INTER_LANCZOS4)
# resized_blured = cv2.GaussianBlur(resized, (9,9), 0)
threshold, resized_binary = cv2.threshold(resized, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
return resized_binary
In [ ]:
num_samples = 0
In [ ]:
def generate_batch(orig_data, forg_data, batch_size = 32):
global num_samples
while True:
orig_pairs = []
forg_pairs = []
gen_gen_labels = []
gen_for_labels = []
all_pairs = []
all_labels = []
# Here we create pairs of Genuine-Genuine image names and Genuine-Forged image names
# For every person we have 24 genuine signatures, hence we have
# 24 choose 2 = 276 Genuine-Genuine image pairs for one person.
# To make Genuine-Forged pairs, we pair every Genuine signature of a person
# with 12 randomly sampled Forged signatures of the same person.
# Thus we make 24 * 12 = 288 Genuine-Forged image pairs for one person.
# In all we have 120 person's data in the training data.
# Total no. of Genuine-Genuine pairs = 160 * 276 = 44160
# Total number of Genuine-Forged pairs = 160 * 288 = 46080
# Total no. of data points = 44160 + 46080 = 90240
for orig, forg in zip(orig_data, forg_data):
orig_pairs.extend(list(itertools.combinations(orig, 2)))
for i in range(len(forg)):
forg_pairs.extend(list(itertools.product(orig[i:i+1], random.sample(forg, len(forg)))))
# Label for Genuine-Genuine pairs is 1
# Label for Genuine-Forged pairs is 0
gen_gen_labels = [1]*len(orig_pairs)
gen_for_labels = [0]*len(forg_pairs)
# Concatenate all the pairs together along with their labels and shuffle them
all_pairs = orig_pairs + forg_pairs
all_labels = gen_gen_labels + gen_for_labels
del orig_pairs, forg_pairs, gen_gen_labels, gen_for_labels
all_pairs, all_labels = shuffle(all_pairs, all_labels)
# print(len(all_pairs))
num_samples = len(all_pairs)
# pairss = all_pairs
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
for ix, pair in enumerate(all_pairs):
img1 = preprocessor_img(pair[0], (img_w, img_h))
img2 = preprocessor_img(pair[1], (img_w, img_h))
# img1 = cv2.imread(pair[0],0)
# img2 = cv2.imread(pair[1],0)
# img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
# img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
img1 = np.atleast_3d(img1)
img2 = np.atleast_3d(img2)
# img2 = img2[...,np.newaxis]
pairs[0][k, :, :, :] = img1
pairs[1][k, :, :, :] = img2
targets[k] = all_labels[ix]
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
visualize generate_batch¶
In [ ]:
ff = generate_batch(orig_groups, forg_groups, batch_size = 32)
pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
num_samples
Out[ ]:
3540
In [ ]:
print(pairs.shape)
print(targets.shape)
(2, 32, 150, 300, 1) (32,)
In [ ]:
print(pairs[0].shape)
(32, 150, 300, 1)
In [ ]:
print(targets)
[0. 0. 1. 1. 0. 0. 0. 0. 0. 0. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 1. 1. 1. 0.]
In [ ]:
i = 2
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
1.0
evaluate¶
In [ ]:
batch_sz = 32
size = len(orig_groups)
num_test_samples = 66*10 + 60*10
num_test_samples = 3540
num_test_samples, num_test_samples//batch_sz, size, batch_sz
Out[ ]:
(3540, 110, 10, 32)
In [ ]:
# dutch
ev = siamese_net.evaluate(
generate_batch(orig_groups, forg_groups, batch_size = batch_sz),
steps = num_test_samples//batch_sz,
)
ev
110/110 [==============================] - 91s 831ms/step - loss: 0.2614 - accuracy: 0.8949
Out[ ]:
[0.26138392090797424, 0.8948863744735718]
Data 6 -- Test kaggle¶
In [ ]:
img_h, img_w, img_ch = 150, 300, 1
image_shape = (img_h, img_w, img_ch)
image_shape
Out[ ]:
(150, 300, 1)
In [ ]:
path = "/content/kaggle30/"
In [ ]:
dir_list = next(os.walk(path))[1]
dir_list.sort(key = int)
len(dir_list)
Out[ ]:
30
In [ ]:
dir_list = shuffle(dir_list)
In [ ]:
dir_list = dir_list[15:25]
In [ ]:
dir_list[:10]
Out[ ]:
['21', '11', '20', '27', '30', '22', '1', '10', '13', '16']
In [ ]:
orig_groups, forg_groups = [], []
In [ ]:
for i,directory in enumerate(dir_list):
# if i==10:
# break
images = os.listdir(path+directory)
images.sort()
images = [path+directory+'/'+x for x in images]
forg_groups.append(images[:5])
orig_groups.append(images[5:])
In [ ]:
print(len(forg_groups))
print(len(orig_groups))
print(len(forg_groups[0]))
print(len(orig_groups[0]))
30 30 5 5
In [ ]:
forg_lengths = [len(x) for x in forg_groups]
orig_lengths = [len(x) for x in orig_groups]
print(forg_lengths)
print(orig_lengths)
[5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5] [5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5]
In [ ]:
# orig_groups[40], forg_groups[40]
In [ ]:
orig_groups, forg_groups = shuffle(orig_groups, forg_groups)
In [ ]:
orig_lengths = [len(x) for x in orig_groups]
forg_lengths = [len(x) for x in forg_groups]
print(orig_lengths)
print(forg_lengths)
[24, 24, 24, 12, 5, 24, 5, 24, 5, 5, 24, 24, 24, 5, 24, 24, 12, 5, 24, 12, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 12, 24, 5, 5, 24, 24, 12, 5, 24, 24, 24, 24, 24, 24, 5, 5, 24, 24, 5, 24, 5, 5, 24, 24, 12, 24, 5, 24, 24, 5, 24, 12, 12, 24, 24, 24, 24, 24, 24, 5, 24, 5, 24, 24, 5, 5, 5, 24, 24, 24, 24, 24, 12, 5, 5, 5, 24, 24, 24, 5, 24, 5, 24, 5, 24, 5, 12, 24, 24, 24, 5, 24, 24, 24, 24, 5, 24, 24] [30, 24, 24, 24, 5, 24, 5, 30, 5, 5, 24, 30, 30, 5, 30, 30, 24, 5, 30, 24, 30, 24, 30, 30, 30, 24, 30, 30, 30, 24, 24, 30, 24, 24, 5, 5, 30, 24, 24, 5, 30, 24, 30, 24, 30, 24, 5, 5, 30, 30, 5, 30, 5, 5, 30, 24, 24, 30, 5, 30, 24, 5, 24, 24, 24, 24, 30, 24, 24, 30, 24, 5, 30, 5, 30, 30, 5, 5, 5, 24, 30, 30, 30, 24, 24, 5, 5, 5, 30, 24, 24, 5, 24, 5, 30, 5, 24, 5, 24, 30, 30, 24, 5, 24, 24, 30, 30, 5, 24, 30]
In [ ]:
orig_train, orig_val = orig_groups[:90], orig_groups[90:]
forg_train, forg_val = forg_groups[:90], forg_groups[90:]
In [ ]:
len(orig_train), len(orig_val), len(forg_train), len(forg_val)
Out[ ]:
(90, 20, 90, 20)
In [ ]:
i = 3
orig_groups[i], forg_groups[i]
Out[ ]:
(['/content/dutch/09/13_026.PNG', '/content/dutch/09/14_026.PNG', '/content/dutch/09/15_026.PNG', '/content/dutch/09/16_026.PNG', '/content/dutch/09/17_026.PNG', '/content/dutch/09/18_026.PNG', '/content/dutch/09/19_026.PNG', '/content/dutch/09/20_026.PNG', '/content/dutch/09/21_026.PNG', '/content/dutch/09/22_026.PNG', '/content/dutch/09/23_026.PNG', '/content/dutch/09/24_026.PNG'], ['/content/dutch/09/01_0119026.PNG', '/content/dutch/09/01_0123026.PNG', '/content/dutch/09/01_0125026.PNG', '/content/dutch/09/01_026.png', '/content/dutch/09/02_0119026.PNG', '/content/dutch/09/02_0123026.PNG', '/content/dutch/09/02_0125026.PNG', '/content/dutch/09/02_026.png', '/content/dutch/09/03_0119026.PNG', '/content/dutch/09/03_0123026.PNG', '/content/dutch/09/03_0125026.PNG', '/content/dutch/09/03_026.png', '/content/dutch/09/04_0119026.PNG', '/content/dutch/09/04_0123026.PNG', '/content/dutch/09/04_0125026.PNG', '/content/dutch/09/04_026.png', '/content/dutch/09/05_026.png', '/content/dutch/09/06_026.png', '/content/dutch/09/07_026.png', '/content/dutch/09/08_026.png', '/content/dutch/09/09_026.png', '/content/dutch/09/10_026.png', '/content/dutch/09/11_026.png', '/content/dutch/09/12_026.png'])
In [ ]:
def visualize_sample_signature(choose=False,idx=0):
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize = (14, 7))
if choose:
k = idx
else:
k = np.random.randint(len(orig_groups))
print(k)
orig_img_names = random.sample(orig_groups[k], 2)
forg_img_name = random.sample(forg_groups[k], 1)
print(orig_img_names+forg_img_name)
orig_img1 = cv2.imread(orig_img_names[0],0)
orig_img2 = cv2.imread(orig_img_names[1],0)
forg_img = cv2.imread(forg_img_name[0],0)
orig_img1 = cv2.resize(orig_img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
orig_img2 = cv2.resize(orig_img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
forg_img = cv2.resize(forg_img, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
ax1.imshow(orig_img1, cmap = 'gray')
ax2.imshow(orig_img2, cmap = 'gray')
ax3.imshow(forg_img, cmap = 'gray')
ax1.set_title('Genuine Copy')
ax1.set_xlabel(1)
# ax1.axis('off')
ax2.set_title('Genuine Copy')
ax2.set_xlabel(2)
# ax2.axis('off')
ax3.set_title('Forged Copy')
ax3.set_xlabel(3)
# ax3.axis('off')
plt.tight_layout()
plt.show()
In [ ]:
visualize_sample_signature(choose=True,idx=16)
16 ['/content/kaggle30/14/g (4).png', '/content/kaggle30/14/g (3).png', '/content/kaggle30/14/f (1).png']
In [ ]:
# list(itertools.combinations(orig_groups[0], 2))
len(list(itertools.combinations(orig_groups[0], 2)))
Out[ ]:
10
In [ ]:
# list(itertools.combinations(orig_groups[2], 2))
len(list(itertools.combinations(orig_groups[2], 2)))
Out[ ]:
10
In [ ]:
# list(itertools.combinations(orig_groups[5], 2))
len(list(itertools.combinations(orig_groups[5], 2)))
Out[ ]:
10
In [ ]:
orig_groups[0][0:1]
Out[ ]:
['/content/kaggle30/1/g (1).png', '/content/kaggle30/1/g (2).png', '/content/kaggle30/1/g (3).png', '/content/kaggle30/1/g (4).png', '/content/kaggle30/1/g (5).png']
In [ ]:
i = 16
img_num=0
list(itertools.product(orig_groups[i][img_num:img_num+1], random.sample(forg_groups[i], len(forg_groups[i]))))
Out[ ]:
[('/content/dutch/05/13_020.PNG', '/content/dutch/05/05_020.png'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/04_0105020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/01_0117020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/02_0127020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/04_020.png'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/01_020.png'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/03_0105020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/03_0213020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/03_0117020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/04_0213020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/02_0213020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/02_020.png'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/02_0101020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/03_0101020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/03_0127020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/01_0213020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/02_0105020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/02_0117020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/04_0101020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/03_020.png'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/04_0117020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/01_0127020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/04_0127020.PNG'),
('/content/dutch/05/13_020.PNG', '/content/dutch/05/01_0105020.PNG')]
In [ ]:
random.sample(forg_groups[0], 5)
Out[ ]:
['/content/kaggle30/1/f (2).png', '/content/kaggle30/1/f (3).png', '/content/kaggle30/1/f (5).png', '/content/kaggle30/1/f (1).png', '/content/kaggle30/1/f (4).png']
In [ ]:
random.sample(forg_groups[0],len(forg_groups[0]))
Out[ ]:
['/content/BHSig260/Hindi/035/H-S-035-F-08.tif', '/content/BHSig260/Hindi/035/H-S-035-F-11.tif', '/content/BHSig260/Hindi/035/H-S-035-F-14.tif', '/content/BHSig260/Hindi/035/H-S-035-F-02.tif', '/content/BHSig260/Hindi/035/H-S-035-F-18.tif', '/content/BHSig260/Hindi/035/H-S-035-F-26.tif', '/content/BHSig260/Hindi/035/H-S-035-F-28.tif', '/content/BHSig260/Hindi/035/H-S-035-F-09.tif', '/content/BHSig260/Hindi/035/H-S-035-F-19.tif', '/content/BHSig260/Hindi/035/H-S-035-F-22.tif', '/content/BHSig260/Hindi/035/H-S-035-F-06.tif', '/content/BHSig260/Hindi/035/H-S-035-F-13.tif', '/content/BHSig260/Hindi/035/H-S-035-F-01.tif', '/content/BHSig260/Hindi/035/H-S-035-F-24.tif', '/content/BHSig260/Hindi/035/H-S-035-F-15.tif', '/content/BHSig260/Hindi/035/H-S-035-F-10.tif', '/content/BHSig260/Hindi/035/H-S-035-F-05.tif', '/content/BHSig260/Hindi/035/H-S-035-F-12.tif', '/content/BHSig260/Hindi/035/H-S-035-F-16.tif', '/content/BHSig260/Hindi/035/H-S-035-F-20.tif', '/content/BHSig260/Hindi/035/H-S-035-F-23.tif', '/content/BHSig260/Hindi/035/H-S-035-F-25.tif', '/content/BHSig260/Hindi/035/H-S-035-F-04.tif', '/content/BHSig260/Hindi/035/H-S-035-F-27.tif', '/content/BHSig260/Hindi/035/H-S-035-F-07.tif', '/content/BHSig260/Hindi/035/H-S-035-F-03.tif', '/content/BHSig260/Hindi/035/H-S-035-F-17.tif', '/content/BHSig260/Hindi/035/H-S-035-F-29.tif', '/content/BHSig260/Hindi/035/H-S-035-F-30.tif', '/content/BHSig260/Hindi/035/H-S-035-F-21.tif']
In [ ]:
print(np.random.randint(0,3))
print(random.randint(0,3))
2 3
In [ ]:
kernel = np.ones((9,9),np.uint8) # default
def preprocessor_img(path, image_shape):
image = cv2.imread(path,0)
blured = cv2.GaussianBlur(image, (9,9), 0)
threshold, binary = cv2.threshold(blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel, iterations=30)
contours, hierarchies = cv2.findContours(closing, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
the_biggest_contour_by_area = max(contours, key=cv2.contourArea)
x,y,w,h = cv2.boundingRect(the_biggest_contour_by_area)
cropped = image[y:y+h, x:x+w]
resized = cv2.resize(cropped, image_shape, interpolation=cv2.INTER_LANCZOS4)
# resized_blured = cv2.GaussianBlur(resized, (9,9), 0)
threshold, resized_binary = cv2.threshold(resized, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
return resized_binary
In [ ]:
num_samples = 0
In [ ]:
def generate_batch(orig_data, forg_data, batch_size = 32):
global num_samples
while True:
orig_pairs = []
forg_pairs = []
gen_gen_labels = []
gen_for_labels = []
all_pairs = []
all_labels = []
# Here we create pairs of Genuine-Genuine image names and Genuine-Forged image names
# For every person we have 24 genuine signatures, hence we have
# 24 choose 2 = 276 Genuine-Genuine image pairs for one person.
# To make Genuine-Forged pairs, we pair every Genuine signature of a person
# with 12 randomly sampled Forged signatures of the same person.
# Thus we make 24 * 12 = 288 Genuine-Forged image pairs for one person.
# In all we have 120 person's data in the training data.
# Total no. of Genuine-Genuine pairs = 160 * 276 = 44160
# Total number of Genuine-Forged pairs = 160 * 288 = 46080
# Total no. of data points = 44160 + 46080 = 90240
for orig, forg in zip(orig_data, forg_data):
orig_pairs.extend(list(itertools.combinations(orig, 2)))
for i in range(len(forg)):
forg_pairs.extend(list(itertools.product(orig[i:i+1], random.sample(forg, len(forg)))))
# Label for Genuine-Genuine pairs is 1
# Label for Genuine-Forged pairs is 0
gen_gen_labels = [1]*len(orig_pairs)
gen_for_labels = [0]*len(forg_pairs)
# Concatenate all the pairs together along with their labels and shuffle them
all_pairs = orig_pairs + forg_pairs
all_labels = gen_gen_labels + gen_for_labels
del orig_pairs, forg_pairs, gen_gen_labels, gen_for_labels
all_pairs, all_labels = shuffle(all_pairs, all_labels)
# print(len(all_pairs))
num_samples = len(all_pairs)
# pairss = all_pairs
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
for ix, pair in enumerate(all_pairs):
img1 = preprocessor_img(pair[0], (img_w, img_h))
img2 = preprocessor_img(pair[1], (img_w, img_h))
# img1 = cv2.imread(pair[0],0)
# img2 = cv2.imread(pair[1],0)
# img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
# img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
img1 = np.atleast_3d(img1)
img2 = np.atleast_3d(img2)
# img2 = img2[...,np.newaxis]
pairs[0][k, :, :, :] = img1
pairs[1][k, :, :, :] = img2
targets[k] = all_labels[ix]
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
visualize generate_batch¶
In [ ]:
ff = generate_batch(orig_groups, forg_groups, batch_size = 32)
# ff = generate_batch(orig_train, forg_train, batch_size = 32)
# ff = generate_batch(orig_val, forg_val, batch_size = 32)
pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
num_samples
Out[ ]:
1050
In [ ]:
num_samples
Out[ ]:
12504
In [ ]:
print(pairs.shape)
print(targets.shape)
(2, 32, 150, 300, 1) (32,)
In [ ]:
print(pairs[0].shape)
(32, 150, 300, 1)
In [ ]:
print(targets)
[1. 0. 1. 0. 1. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 1. 1. 1. 0. 0. 0.]
In [ ]:
i = 1
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
0.0
In [ ]:
evaluate¶
In [ ]:
batch_sz = 32
size = len(orig_groups)
num_test_samples = 10*30 + 25*30
num_test_samples = 1050
num_test_samples, num_test_samples//batch_sz, size, batch_sz
Out[ ]:
(1050, 32, 30, 32)
In [ ]:
# kaggle30
ev = siamese_net.evaluate(
generate_batch(orig_groups, forg_groups, batch_size = batch_sz),
steps = num_test_samples//batch_sz,
)
ev
32/32 [==============================] - 59s 2s/step - loss: 0.2600 - accuracy: 0.8994
Out[ ]:
[0.2599816620349884, 0.8994140625]
Model¶
In [ ]:
# image_shape = (pairs.shape[2], pairs.shape[3], pairs.shape[4])
image_shape = (img_h, img_w, img_ch)
image_shape
Out[ ]:
(150, 300, 1)
In [ ]:
from tensorflow.keras.applications import DenseNet201
base_network = DenseNet201(weights=None, include_top=False, input_shape=image_shape)
base_network.summary()
In [ ]:
base_network.trainable = True
In [ ]:
# tf.keras.utils.plot_model( base_network,
# to_file=r"/content/drive/MyDrive/graduation project (ML)/model_5/base_network.png",
# show_shapes=True,
# show_layer_names=True,
# show_dtype=True)
In [ ]:
# Define the tensors for the two input images
left_input = Input(image_shape)
right_input = Input(image_shape)
# Convolutional Neural Network
model = Sequential()
model.add(base_network)
model.add(Flatten())
# Generate the encodings (feature vectors) for the two images
encoded_l = model(left_input)
encoded_r = model(right_input)
# Add a customized layer to compute the absolute difference between the encodings
merge_layer = Lambda(lambda tensors:K.abs(tensors[0] - tensors[1]))([encoded_l, encoded_r])
# Add a dense layer with a sigmoid unit to generate the similarity score
prediction = Dense(1,activation='sigmoid')(merge_layer)
# Connect the inputs with the outputs
siamese_net = Model(inputs=[left_input,right_input],outputs=prediction)
siamese_net.summary()
In [ ]:
# tf.keras.utils.plot_model(siamese_net,
# to_file=r"/content/drive/MyDrive/graduation project (ML)/model_5/siamese_net.png",
# show_shapes=True,
# show_layer_names=True,
# show_dtype=True)
In [ ]:
callbacks = [
EarlyStopping(patience=15, restore_best_weights=True),
CSVLogger(r"/content/drive/MyDrive/graduation project (ML)/model_5/model_history_log.csv", append=True),
# ReduceLROnPlateau(factor=0.1, patience=5, min_lr=0.000001),
ModelCheckpoint(r"/content/drive/MyDrive/graduation project (ML)/model_5/signet-{epoch:003d}.h5"),
# ModelCheckpoint(r"/content/drive/MyDrive/graduation project (ML)/model_5/signet_Best_Model.h5", save_best_only=True)
]
In [ ]:
batch_sz = 32
num_train_samples = 276*120 + 288*120
num_val_samples = num_test_samples = 276*20 + 288*20
num_train_samples, num_val_samples, num_test_samples, num_train_samples//batch_sz
Out[ ]:
(67680, 11280, 11280, 2115)
In [ ]:
# rms = RMSprop(lr=1e-4, rho=0.9, epsilon=1e-08)
adam = Adam(lr=1e-6)
# rms = RMSprop()
# adam = Adam()
siamese_net.compile(loss = "binary_crossentropy", optimizer='adam', metrics=["accuracy"])
In [ ]:
new_history = siamese_net.fit( generate_batch(orig_train, forg_train, batch_sz),
steps_per_epoch = num_train_samples//batch_sz,
epochs=3,
validation_data = generate_batch(orig_val, forg_val, batch_sz),
validation_steps = num_val_samples//batch_sz,
callbacks = callbacks
)
In [ ]:
siamese_net.save(r"/content/drive/MyDrive/graduation project (ML)/model_5/model_3epoch")
In [ ]:
siamese_net.save(r"/content/drive/MyDrive/graduation project (ML)/model_5/model_3epoch.h5")
In [ ]:
In [ ]:
In [ ]:
In [ ]:
siamese_net = keras.models.load_model(r"/content/drive/MyDrive/graduation project (ML)/model_5/Train 11/model_34.h5")
In [ ]:
siamese_net.summary()
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_2 (InputLayer) [(None, 150, 300, 1) 0
__________________________________________________________________________________________________
input_3 (InputLayer) [(None, 150, 300, 1) 0
__________________________________________________________________________________________________
sequential (Sequential) (None, 69120) 18315712 input_2[0][0]
input_3[0][0]
__________________________________________________________________________________________________
lambda (Lambda) (None, 69120) 0 sequential[0][0]
sequential[1][0]
__________________________________________________________________________________________________
dense (Dense) (None, 1) 69121 lambda[0][0]
==================================================================================================
Total params: 18,384,833
Trainable params: 18,155,777
Non-trainable params: 229,056
__________________________________________________________________________________________________
In [ ]:
callbacks = [
EarlyStopping(patience=15, restore_best_weights=True),
CSVLogger(r"/content/drive/MyDrive/graduation project (ML)/model_5/model_history_log.csv", append=True),
ModelCheckpoint(r"/content/drive/MyDrive/graduation project (ML)/model_5/signet-{epoch:003d}.h5"),
# ReduceLROnPlateau(factor=0.1, patience=5, min_lr=0.000001),
# ModelCheckpoint(r"/content/drive/MyDrive/graduation project (ML)/model_5/signet_Best_Model.h5", save_best_only=True)
]
In [ ]:
batch_sz = 32
num_train_samples = 276*len(orig_train) + 288*len(orig_train)
num_val_samples = num_test_samples = 276*len(orig_val) + 288*len(orig_val)
num_train_samples, num_val_samples, num_test_samples, num_train_samples//batch_sz, num_val_samples//batch_sz, len(orig_train), len(orig_val)
Out[ ]:
(14100, 2820, 2820, 440, 88, 25, 5)
In [ ]:
# kaggle
batch_sz = 32
num_train_samples = 10*len(orig_train) + 25*len(orig_train)
num_val_samples = num_test_samples = 10*len(orig_val) + 25*len(orig_val)
num_train_samples, num_val_samples, num_test_samples, num_train_samples//batch_sz
Out[ ]:
(875, 175, 175, 27)
In [ ]:
# kaggle + dutch + cedar
batch_sz = 32
num_train_samples = 4221
num_val_samples = num_test_samples = 1349
num_train_samples, num_val_samples, num_test_samples, num_train_samples//batch_sz
Out[ ]:
(4221, 1349, 1349, 131)
In [ ]:
# kaggle + dutch + cedar + ban + hindi
batch_sz = 32
num_train_samples = 57486
num_val_samples = num_test_samples = 12504
num_train_samples, num_val_samples, num_train_samples//batch_sz, num_val_samples//batch_sz
Out[ ]:
(57486, 12504, 1796, 390)
In [ ]:
adam = Adam(lr=1e-6)
siamese_net.compile(loss = "binary_crossentropy", optimizer='adam', metrics=["accuracy"])
In [ ]:
new_history = siamese_net.fit( generate_batch(orig_train, forg_train, batch_sz),
steps_per_epoch = num_train_samples//batch_sz,
epochs=1,
validation_data = generate_batch(orig_val, forg_val, batch_sz),
validation_steps = num_val_samples//batch_sz,
callbacks = callbacks
)
1796/1796 [==============================] - 2112s 1s/step - loss: 0.0727 - accuracy: 0.9743 - val_loss: 0.5458 - val_accuracy: 0.8080
In [ ]:
siamese_net.save(r"/content/drive/MyDrive/graduation project (ML)/model_5/model_30epoch.h5")
In [ ]:
siamese_net.save(r"/content/drive/MyDrive/graduation project (ML)/model_5/model_30epoch")
INFO:tensorflow:Assets written to: /content/drive/MyDrive/graduation project (ML)/model_5/model_16epoch/assets
In [ ]:
# keras.backend.clear_session()
# del siamese_net
In [ ]:
history = new_history
In [ ]:
print(siamese_net.metrics_names)
print(history.params)
print(history.history.keys())
['loss', 'accuracy']
{'verbose': 1, 'epochs': 3, 'steps': 2115}
dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])
In [ ]:
pd.DataFrame(history.history).plot(figsize=(10, 5))
plt.grid(True)
# plt.gca().set_ylim(0, 1)
plt.gca()
plt.show()
In [ ]:
pd.DataFrame(history.history).plot(figsize=(16, 16), layout=(4,4), subplots=True)
plt.tight_layout()
plt.show()
ADD history¶
In [ ]:
history = {'loss':[],'accuracy':[],'val_loss':[], 'val_accuracy':[]}
In [ ]:
from copy import copy
def addNewHistory(old_history,new_history):
history = copy(old_history)
for i in old_history.keys():
# print(i)
history[i]+=new_history.history[i]
return history
In [ ]:
history = addNewHistory(history,new_history)
In [ ]:
pickle.dump((history), open(r"/content/drive/MyDrive/graduation project (ML)/model_5/history_3epoch.pickle", "wb") )
In [ ]:
history = pickle.load(open(r"/content/drive/MyDrive/graduation project (ML)/model_5/history_3epoch.pickle", "rb"))
In [ ]:
pd.DataFrame(history).plot(figsize=(10, 5))
plt.grid(True)
# plt.gca().set_ylim(0, 1)
plt.gca()
plt.show()
In [ ]:
pd.DataFrame(history).plot(figsize=(16, 16), layout=(4,4), subplots=True)
plt.tight_layout()
plt.show()
In [ ]:
In [ ]:
history = pickle.load(open(r"/content/drive/MyDrive/graduation project (ML)/model_5/history.pickle", "rb"))
In [ ]:
history = addNewHistory(history,new_history)
In [ ]:
pickle.dump((history), open(r"/content/drive/MyDrive/graduation project (ML)/model_5/history_10epoch.pickle", "wb") )
In [ ]:
pd.DataFrame(history).plot(figsize=(10, 5))
plt.grid(True)
plt.gca()
plt.show()
In [ ]:
pd.DataFrame(history).plot(figsize=(16, 16), layout=(4,4), subplots=True)
plt.tight_layout()
plt.show()
Load Model for Test¶
In [ ]:
siamese_net = keras.models.load_model("models/model_3/model_3_5epoch_rms.h5")
Test 1¶
In [ ]:
# ev = siamese_net.evaluate([pairs_test[:,0], pairs_test[:,1]], labels_test[:], steps=50)
ev = siamese_net.evaluate([pairs_train[:,0], pairs_train[:,1]], labels_train[:])
ev
540/540 [==============================] - 128s 226ms/step - loss: 0.1745 - accuracy: 0.9545
Out[ ]:
[0.17453065514564514, 0.9545139074325562]
In [ ]:
path = "dataset/test/"
In [ ]:
X,y,c=loadimgs(path,limit=True,limit_number=7)
--------------------------------------------------------------------------- NameError Traceback (most recent call last) <ipython-input-28-7eefae8635c8> in <module> ----> 1 X,y,c=loadimgs(path,limit=True,limit_number=7) NameError: name 'loadimgs' is not defined
In [ ]:
print(X.shape)
print(y.shape)
(7, 54, 100, 200, 3) (378,)
In [ ]:
pairs_test, labels_test = make_pairs(X, y)
In [ ]:
print(pairs_test.shape)
print(labels_test.shape)
(756, 2, 100, 200, 3) (756,)
In [ ]:
ev = siamese_net.evaluate([pairs_test[:,0], pairs_test[:,1]], labels_test[:])
ev
24/24 [==============================] - 6s 227ms/step - loss: 0.7266 - accuracy: 0.7222
Out[ ]:
[0.7266002297401428, 0.7222222089767456]
In [ ]:
i = 266
print(labels_test[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs_test[i][0],'gray')
plt.subplot(132)
plt.imshow(pairs_test[i][1],'gray')
plt.tight_layout()
plt.show()
1
In [ ]:
img_1 = pairs_test[i][0].copy()
img_1 = img_1.reshape(1,pairs_test[i][0].shape[0],pairs_test[i][0].shape[1],3)
img_1.shape
img_2 = pairs_test[i][1].copy()
img_2 = img_2.reshape(1,pairs_test[i][1].shape[0],pairs_test[i][1].shape[1],3)
img_2.shape
Out[ ]:
(1, 100, 200, 3)
In [ ]:
prediction_prob = siamese_net.predict([img_1, img_2])
print(prediction_prob)
print(prediction_prob.round())
[[0.5346107]] [[1.]]
Test 2¶
Test 2¶
In [ ]:
path = "dataset/test/"
In [ ]:
dir_list = os.listdir(path)
len(dir_list)
Out[ ]:
100
In [ ]:
data = []
for directory in dir_list:
images = os.listdir(path+directory)
images = [path+directory+'/'+x for x in images]
data.append(images[30:])
In [ ]:
len(data[0])
Out[ ]:
24
In [ ]:
data[6][:]
Out[ ]:
['dataset/test/007/B-S-7-G-01.tif', 'dataset/test/007/B-S-7-G-02.tif', 'dataset/test/007/B-S-7-G-03.tif', 'dataset/test/007/B-S-7-G-04.tif', 'dataset/test/007/B-S-7-G-05.tif', 'dataset/test/007/B-S-7-G-06.tif', 'dataset/test/007/B-S-7-G-07.tif', 'dataset/test/007/B-S-7-G-08.tif', 'dataset/test/007/B-S-7-G-09.tif', 'dataset/test/007/B-S-7-G-10.tif', 'dataset/test/007/B-S-7-G-11.tif', 'dataset/test/007/B-S-7-G-12.tif', 'dataset/test/007/B-S-7-G-13.tif', 'dataset/test/007/B-S-7-G-14.tif', 'dataset/test/007/B-S-7-G-15.tif', 'dataset/test/007/B-S-7-G-16.tif', 'dataset/test/007/B-S-7-G-17.tif', 'dataset/test/007/B-S-7-G-18.tif', 'dataset/test/007/B-S-7-G-19.tif', 'dataset/test/007/B-S-7-G-20.tif', 'dataset/test/007/B-S-7-G-21.tif', 'dataset/test/007/B-S-7-G-22.tif', 'dataset/test/007/B-S-7-G-23.tif', 'dataset/test/007/B-S-7-G-24.tif']
In [ ]:
img_h, img_w, img_ch = 100, 200 , 3
In [ ]:
visualize_sample_signature(choose=True,idx=15)
15 ['dataset/test/016/B-S-16-G-05.tif', 'dataset/test/016/B-S-16-G-03.tif', 'dataset/test/016/B-S-16-G-20.tif']
In [ ]:
ff = generate_batch(data, batch_size = 32)
pairs, targets = next(ff)
pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
print(pairs.shape)
print(targets.shape)
(2, 32, 100, 200, 3) (32,)
In [ ]:
print(targets)
[1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0.]
In [ ]:
i = 3
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
0.0
In [ ]:
batch_sz = 32
num_test_samples = len(data) * len(data[0]) * 2
# steps_per_epoch = num_train_samples//batch_sz,
num_test_samples
Out[ ]:
4800
In [ ]:
ev = siamese_net.evaluate(generate_batch(data, batch_size = batch_sz), steps = num_test_samples//batch_sz,)
ev
150/150 [==============================] - 179s 1s/step - loss: 0.5094 - accuracy: 0.8050
Out[ ]:
[0.5094283223152161, 0.8050000071525574]
Test 2 another way¶
In [ ]:
i = 1
print(targets[i])
img_1 = pairs[0][i].copy()
print(img_1.shape)
img_1 = img_1.reshape(1, pairs[0][i].shape[0], pairs[0][i].shape[1], pairs[0][i].shape[2])
print(img_1.shape)
img_2 = pairs[1][i].copy()
print(img_2.shape)
img_2 = img_2.reshape(1, pairs[0][i].shape[0], pairs[0][i].shape[1], pairs[0][i].shape[2])
print(img_2.shape)
0.0 (100, 200, 3) (1, 100, 200, 3) (100, 200, 3) (1, 100, 200, 3)
In [ ]:
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
0.0
In [ ]:
prediction_prob = siamese_net.predict([img_1, img_2])
print(prediction_prob)
print(prediction_prob.round())
print(prediction_prob[0][0].round())
[[0.6091775]] [[1.]] 1.0
In [ ]:
In [ ]:
d = generate_batch(data, batch_size = 1)
pairs, targets = next(d)
pairs, targets = next(d)
pairs, targets = next(d)
print(np.array(pairs).shape)
i = 0
print(targets[i])
img_1 = pairs[0].copy()
print(img_1.shape)
# img_2 = pairs[1][i].copy()
# print(img_2.shape)
img_2 = pairs[1].copy()
print(img_2.shape)
(2, 1, 100, 200, 3) 1.0 (1, 100, 200, 3) (1, 100, 200, 3)
In [ ]:
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
1.0
In [ ]:
prediction_prob = siamese_net.predict([pairs[0], pairs[1]])
print(prediction_prob)
print(prediction_prob.round())
print(prediction_prob[0][0].round())
[[0.7506307]] [[1.]] 1.0
Test 2 another way¶
In [ ]:
d = generate_batch(data, batch_size = 5)
pairs, targets = next(d)
In [ ]:
prediction_prob = siamese_net.predict([pairs[0], pairs[1]])
print(prediction_prob)
print(prediction_prob.round())
print(prediction_prob[0][0].round())
print(prediction_prob.round().flatten())
[[0.86238354] [0.11163066] [0.58783156] [0.00089584] [0.8684073 ]] [[1.] [0.] [1.] [0.] [1.]] 1.0 [1. 0. 1. 0. 1.]
Test 2 another way¶
In [ ]:
num_test_samples = len(data) * len(data[0]) * 2
num_test_samples
Out[ ]:
4800
In [ ]:
# test_gen = generate_batch(data, batch_size = 1)
test_gen = generate_batch(data, batch_size = 100)
In [ ]:
(img1, img2), label = next(test_gen)
print(img1.shape)
print(img2.shape)
print(label)
(1, 100, 200, 3) (1, 100, 200, 3) [1.]
In [ ]:
# %%time
# y_true, y_pred = [], []
# for i in range(num_test_samples):
# (img1, img2), label = next(test_gen)
# y_true.append(label)
# y_pred.append(siamese_net.predict([img1, img2])[0][0].round())
In [ ]:
%%time
y_true, y_pred = [], []
for i in range(num_test_samples//100):
(img1, img2), label = next(test_gen)
y_true.extend(label.tolist())
y_pred.extend(siamese_net.predict([img1, img2]).round().flatten().tolist())
Wall time: 4min 27s
In [ ]:
CM = confusion_matrix(y_true, y_pred)
print(CM)
[[2230 170] [ 771 1629]]
In [ ]:
sns.heatmap(CM, center = True, annot=True, fmt="d")
plt.show()
In [ ]:
print(classification_report(y_true, y_pred))
precision recall f1-score support
0.0 0.74 0.93 0.83 2400
1.0 0.91 0.68 0.78 2400
accuracy 0.80 4800
macro avg 0.82 0.80 0.80 4800
weighted avg 0.82 0.80 0.80 4800
In [ ]:
test_gen = generate_batch(data, batch_size = 1)
In [ ]:
def predict_score():
test_point, test_label = next(test_gen)
img1, img2 = test_point[0], test_point[1]
print(img1.shape)
print(np.squeeze(img1).shape)
fig, (ax1, ax2) = plt.subplots(1, 2, figsize = (10, 10))
ax1.imshow(np.squeeze(img1), cmap='gray')
ax2.imshow(np.squeeze(img2), cmap='gray')
ax1.set_title('Genuine')
if test_label == 1:
ax2.set_title('Genuine')
else:
ax2.set_title('Forged')
ax1.axis('off')
ax2.axis('off')
plt.show()
result = siamese_net.predict([img1, img2])
diff = result[0][0]
print("Difference Score = ", diff)
In [ ]:
predict_score()
(1, 100, 200, 3) (100, 200, 3)
Difference Score = 0.9002534
Test 3¶
In [ ]:
# siamese_net = keras.models.load_model(r"/content/drive/MyDrive/graduation project (ML)/model_5/Train 5/signet-002.h5")
In [ ]:
img_h, img_w, img_ch = 150, 300, 1
image_shape = (img_h, img_w, img_ch)
image_shape
Out[ ]:
(150, 300, 1)
In [ ]:
path = "/content/BHSig260/Bengali/"
path = "/content/BHSig260/Hindi/"
In [ ]:
dir_list = next(os.walk(path))[1]
dir_list.sort(reverse=True)
# dir_list.sort()
len(dir_list)
Out[ ]:
160
In [ ]:
dir_list = shuffle(dir_list)
In [ ]:
orig_groups, forg_groups = [], []
for i,directory in enumerate(dir_list):
if i==10:
break
images = os.listdir(path+directory)
images.sort()
images = [path+directory+'/'+x for x in images]
forg_groups.append(images[:30])
orig_groups.append(images[30:])
In [ ]:
print(len(orig_groups))
print(len(forg_groups))
print(len(orig_groups[0]))
print(len(forg_groups[0]))
10 10 24 30
In [ ]:
orig_lengths = [len(x) for x in orig_groups]
forg_lengths = [len(x) for x in forg_groups]
print(orig_lengths)
print(forg_lengths)
[24, 24, 24, 24, 24, 24, 24, 24, 24, 24] [30, 30, 30, 30, 30, 30, 30, 30, 30, 30]
In [ ]:
kernel = np.ones((9,9),np.uint8) # default
def preprocessor_img(path, image_shape):
image = cv2.imread(path,0)
blured = cv2.GaussianBlur(image, (9,9), 0)
threshold, binary = cv2.threshold(blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel, iterations=30)
contours, hierarchies = cv2.findContours(closing, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
the_biggest_contour_by_area = max(contours, key=cv2.contourArea)
x,y,w,h = cv2.boundingRect(the_biggest_contour_by_area)
cropped = image[y:y+h, x:x+w]
resized = cv2.resize(cropped, image_shape, interpolation=cv2.INTER_LANCZOS4)
# resized_blured = cv2.GaussianBlur(resized, (9,9), 0)
threshold, resized_binary = cv2.threshold(resized, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
return resized_binary
In [ ]:
num_samples = 0
In [ ]:
def generate_batch(orig_data, forg_data, batch_size = 32):
global num_samples
while True:
orig_pairs = []
forg_pairs = []
gen_gen_labels = []
gen_for_labels = []
all_pairs = []
all_labels = []
# Here we create pairs of Genuine-Genuine image names and Genuine-Forged image names
# For every person we have 24 genuine signatures, hence we have
# 24 choose 2 = 276 Genuine-Genuine image pairs for one person.
# To make Genuine-Forged pairs, we pair every Genuine signature of a person
# with 12 randomly sampled Forged signatures of the same person.
# Thus we make 24 * 12 = 288 Genuine-Forged image pairs for one person.
# In all we have 120 person's data in the training data.
# Total no. of Genuine-Genuine pairs = 160 * 276 = 44160
# Total number of Genuine-Forged pairs = 160 * 288 = 46080
# Total no. of data points = 44160 + 46080 = 90240
for orig, forg in zip(orig_data, forg_data):
orig_pairs.extend(list(itertools.combinations(orig, 2)))
for i in range(len(forg)):
forg_pairs.extend(list(itertools.product(orig[i:i+1], random.sample(forg, len(forg)))))
# Label for Genuine-Genuine pairs is 1
# Label for Genuine-Forged pairs is 0
gen_gen_labels = [1]*len(orig_pairs)
gen_for_labels = [0]*len(forg_pairs)
# Concatenate all the pairs together along with their labels and shuffle them
all_pairs = orig_pairs + forg_pairs
all_labels = gen_gen_labels + gen_for_labels
del orig_pairs, forg_pairs, gen_gen_labels, gen_for_labels
all_pairs, all_labels = shuffle(all_pairs, all_labels)
# print(len(all_pairs))
num_samples = len(all_pairs)
# pairss = all_pairs
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
for ix, pair in enumerate(all_pairs):
img1 = preprocessor_img(pair[0], (img_w, img_h))
img2 = preprocessor_img(pair[1], (img_w, img_h))
# img1 = cv2.imread(pair[0],0)
# img2 = cv2.imread(pair[1],0)
# img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
# img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
img1 = np.atleast_3d(img1)
img2 = np.atleast_3d(img2)
# img2 = img2[...,np.newaxis]
pairs[0][k, :, :, :] = img1
pairs[1][k, :, :, :] = img2
targets[k] = all_labels[ix]
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
visualize generate_batch¶
In [ ]:
ff = generate_batch(orig_groups, forg_groups, batch_size = 32)
# ff = generate_batch(orig_train, forg_train, batch_size = 32)
# ff = generate_batch(orig_val, forg_val, batch_size = 32)
pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
num_samples
Out[ ]:
9960
In [ ]:
print(pairs.shape)
print(targets.shape)
print(pairs[0].shape)
print(targets)
(2, 32, 150, 300, 1) (32,) (32, 150, 300, 1) [0. 0. 0. 0. 1. 0. 1. 0. 1. 0. 0. 1. 1. 0. 0. 0. 0. 1. 0. 0. 0. 0. 1. 1. 0. 0. 0. 0. 0. 0. 0. 1.]
In [ ]:
i = 1
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
0.0
evaluate¶
In [ ]:
batch_sz = 100
size = len(orig_groups)
num_train_samples = 276*size + 288*size
num_val_samples = num_test_samples = 276*size + 288*size
num_train_samples = num_val_samples = num_test_samples = 9960
num_train_samples, num_val_samples, num_test_samples, num_train_samples//batch_sz, size, batch_sz
Out[ ]:
(9960, 9960, 9960, 99, 10, 100)
In [ ]:
# Bengali
ev = siamese_net.evaluate(
generate_batch(orig_groups, forg_groups, batch_size = batch_sz),
steps = num_test_samples//batch_sz,
)
ev
99/99 [==============================] - 206s 2s/step - loss: 0.4900 - accuracy: 0.8210
Out[ ]:
[0.49004265666007996, 0.8210101127624512]
In [ ]:
# Hindi
ev = siamese_net.evaluate(
generate_batch(orig_groups, forg_groups, batch_size = batch_sz),
steps = num_test_samples//batch_sz,
)
ev
99/99 [==============================] - 206s 2s/step - loss: 0.4316 - accuracy: 0.8358
Out[ ]:
[0.4316082298755646, 0.8357575535774231]
Test 4¶
In [ ]:
# siamese_net = keras.models.load_model(r"/content/drive/MyDrive/graduation project (ML)/model_5/Train 4/model_10epoch.h5")
In [ ]:
img_h, img_w, img_ch = 150, 300, 1
image_shape = (img_h, img_w, img_ch)
image_shape
Out[ ]:
(150, 300, 1)
In [ ]:
path = "/content/cedar/"
In [ ]:
dir_list = next(os.walk(path))[1]
# dir_list.sort(key=int,reverse=True)
dir_list.sort(key=int)
len(dir_list)
Out[ ]:
55
In [ ]:
dir_list = shuffle(dir_list)
In [ ]:
dir_list[:10]
Out[ ]:
['21', '36', '15', '13', '54', '30', '46', '11', '43', '7']
In [ ]:
orig_groups, forg_groups = [], []
In [ ]:
for i,directory in enumerate(dir_list):
if i==10:
break
images = os.listdir(path+directory)
images.sort()
images = [path+directory+'/'+x for x in images]
forg_groups.append(images[:24])
orig_groups.append(images[24:])
In [ ]:
orig_lengths = [len(x) for x in orig_groups]
forg_lengths = [len(x) for x in forg_groups]
print(orig_lengths)
print(forg_lengths)
[24, 24, 24, 24, 24, 24, 24, 24, 24, 24] [24, 24, 24, 24, 24, 24, 24, 24, 24, 24]
In [ ]:
orig_groups, forg_groups = shuffle(orig_groups, forg_groups)
In [ ]:
orig_lengths = [len(x) for x in orig_groups]
forg_lengths = [len(x) for x in forg_groups]
print(orig_lengths)
print(forg_lengths)
[24, 24, 24, 24, 24, 24, 24, 24, 24, 24] [24, 24, 24, 24, 24, 24, 24, 24, 24, 24]
In [ ]:
forg = forg_groups[:10]
orig = orig_groups[:10]
In [ ]:
# orig_train, orig_val = orig_groups[:45], orig_groups[45:]
# forg_train, forg_val = forg_groups[:45], forg_groups[45:]
In [ ]:
len(orig_groups), len(forg_groups), len(orig_groups[0]), len(forg_groups[0])
Out[ ]:
(10, 10, 24, 24)
In [ ]:
orig_groups[0], forg_groups[0]
Out[ ]:
(['/content/cedar/19/original_19_1.png', '/content/cedar/19/original_19_10.png', '/content/cedar/19/original_19_11.png', '/content/cedar/19/original_19_12.png', '/content/cedar/19/original_19_13.png', '/content/cedar/19/original_19_14.png', '/content/cedar/19/original_19_15.png', '/content/cedar/19/original_19_16.png', '/content/cedar/19/original_19_17.png', '/content/cedar/19/original_19_18.png', '/content/cedar/19/original_19_19.png', '/content/cedar/19/original_19_2.png', '/content/cedar/19/original_19_20.png', '/content/cedar/19/original_19_21.png', '/content/cedar/19/original_19_22.png', '/content/cedar/19/original_19_23.png', '/content/cedar/19/original_19_24.png', '/content/cedar/19/original_19_3.png', '/content/cedar/19/original_19_4.png', '/content/cedar/19/original_19_5.png', '/content/cedar/19/original_19_6.png', '/content/cedar/19/original_19_7.png', '/content/cedar/19/original_19_8.png', '/content/cedar/19/original_19_9.png'], ['/content/cedar/19/forgeries_19_1.png', '/content/cedar/19/forgeries_19_10.png', '/content/cedar/19/forgeries_19_11.png', '/content/cedar/19/forgeries_19_12.png', '/content/cedar/19/forgeries_19_13.png', '/content/cedar/19/forgeries_19_14.png', '/content/cedar/19/forgeries_19_15.png', '/content/cedar/19/forgeries_19_16.png', '/content/cedar/19/forgeries_19_17.png', '/content/cedar/19/forgeries_19_18.png', '/content/cedar/19/forgeries_19_19.png', '/content/cedar/19/forgeries_19_2.png', '/content/cedar/19/forgeries_19_20.png', '/content/cedar/19/forgeries_19_21.png', '/content/cedar/19/forgeries_19_22.png', '/content/cedar/19/forgeries_19_23.png', '/content/cedar/19/forgeries_19_24.png', '/content/cedar/19/forgeries_19_3.png', '/content/cedar/19/forgeries_19_4.png', '/content/cedar/19/forgeries_19_5.png', '/content/cedar/19/forgeries_19_6.png', '/content/cedar/19/forgeries_19_7.png', '/content/cedar/19/forgeries_19_8.png', '/content/cedar/19/forgeries_19_9.png'])
In [ ]:
kernel = np.ones((9,9),np.uint8) # default
def preprocessor_img(path, image_shape):
image = cv2.imread(path,0)
blured = cv2.GaussianBlur(image, (9,9), 0)
threshold, binary = cv2.threshold(blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel, iterations=30)
contours, hierarchies = cv2.findContours(closing, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
the_biggest_contour_by_area = max(contours, key=cv2.contourArea)
x,y,w,h = cv2.boundingRect(the_biggest_contour_by_area)
cropped = image[y:y+h, x:x+w]
resized = cv2.resize(cropped, image_shape, interpolation=cv2.INTER_LANCZOS4)
# resized_blured = cv2.GaussianBlur(resized, (9,9), 0)
threshold, resized_binary = cv2.threshold(resized, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
return resized_binary
In [ ]:
num_samples = 0
In [ ]:
def generate_batch(orig_data, forg_data, batch_size = 32):
global num_samples
while True:
orig_pairs = []
forg_pairs = []
gen_gen_labels = []
gen_for_labels = []
all_pairs = []
all_labels = []
# Here we create pairs of Genuine-Genuine image names and Genuine-Forged image names
# For every person we have 24 genuine signatures, hence we have
# 24 choose 2 = 276 Genuine-Genuine image pairs for one person.
# To make Genuine-Forged pairs, we pair every Genuine signature of a person
# with 12 randomly sampled Forged signatures of the same person.
# Thus we make 24 * 12 = 288 Genuine-Forged image pairs for one person.
# In all we have 120 person's data in the training data.
# Total no. of Genuine-Genuine pairs = 160 * 276 = 44160
# Total number of Genuine-Forged pairs = 160 * 288 = 46080
# Total no. of data points = 44160 + 46080 = 90240
for orig, forg in zip(orig_data, forg_data):
orig_pairs.extend(list(itertools.combinations(orig, 2)))
for i in range(len(forg)):
forg_pairs.extend(list(itertools.product(orig[i:i+1], random.sample(forg, len(forg)))))
# Label for Genuine-Genuine pairs is 1
# Label for Genuine-Forged pairs is 0
gen_gen_labels = [1]*len(orig_pairs)
gen_for_labels = [0]*len(forg_pairs)
# Concatenate all the pairs together along with their labels and shuffle them
all_pairs = orig_pairs + forg_pairs
all_labels = gen_gen_labels + gen_for_labels
del orig_pairs, forg_pairs, gen_gen_labels, gen_for_labels
all_pairs, all_labels = shuffle(all_pairs, all_labels)
# print(len(all_pairs))
num_samples = len(all_pairs)
# pairss = all_pairs
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
for ix, pair in enumerate(all_pairs):
img1 = preprocessor_img(pair[0], (img_w, img_h))
img2 = preprocessor_img(pair[1], (img_w, img_h))
# img1 = cv2.imread(pair[0],0)
# img2 = cv2.imread(pair[1],0)
# img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
# img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
img1 = np.atleast_3d(img1)
img2 = np.atleast_3d(img2)
# img2 = img2[...,np.newaxis]
pairs[0][k, :, :, :] = img1
pairs[1][k, :, :, :] = img2
targets[k] = all_labels[ix]
k += 1
if k == batch_size:
# yield np.array(pairs), np.array(targets)
yield pairs, targets
k = 0
pairs=[np.zeros((batch_size, img_h, img_w, img_ch)) for i in range(2)]
targets=np.zeros((batch_size,))
visualize generate_batch¶
In [ ]:
ff = generate_batch(orig_groups, forg_groups, batch_size = 32)
# ff = generate_batch(orig_train, forg_train, batch_size = 32)
# ff = generate_batch(orig_val, forg_val, batch_size = 32)
pairs, targets = next(ff)
pairs, targets = np.array(pairs), np.array(targets)
In [ ]:
num_samples
Out[ ]:
8520
In [ ]:
print(pairs.shape)
print(targets.shape)
print(pairs[0].shape)
print(targets)
(2, 32, 150, 300, 1) (32,) (32, 150, 300, 1) [0. 1. 0. 1. 0. 0. 0. 1. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 0.]
In [ ]:
i = 1
print(targets[i])
plt.figure(figsize = (15, 10))
plt.subplot(131)
plt.imshow(pairs[0][i],'gray')
plt.subplot(132)
plt.imshow(pairs[1][i],'gray')
plt.tight_layout()
plt.show()
1.0
evaluate¶
In [ ]:
batch_sz = 100
size = len(orig_groups)
# size = len(orig)
num_train_samples = 276*size + 288*size
num_val_samples = num_test_samples = 276*size + 288*size
num_train_samples = num_val_samples = num_test_samples = 8520
num_train_samples, num_val_samples, num_test_samples, num_train_samples//batch_sz, size, batch_sz
Out[ ]:
(8520, 8520, 8520, 85, 10, 100)
In [ ]:
# cedar
ev = siamese_net.evaluate(
generate_batch(orig_groups, forg_groups, batch_size = batch_sz),
steps = num_test_samples//batch_sz,
)
# ev = siamese_net.evaluate(
# generate_batch(orig, forg, batch_size = batch_sz),
# steps = num_test_samples//batch_sz,
# )
ev
85/85 [==============================] - 172s 2s/step - loss: 0.1477 - accuracy: 0.9433
Out[ ]:
[0.14774958789348602, 0.943294107913971]
test 5¶
In [ ]:
from platform import python_version
print(python_version())
print()
3.7.10
In [ ]:
!which python
/usr/local/bin/python
In [ ]:
siamese_net = keras.models.load_model(r"/content/drive/MyDrive/graduation project (ML)/model_5/Train 11/model_34.h5")
In [ ]:
# siamese_net.save(r"/content/drive/MyDrive/graduation project (ML)/model_5/Train 11/model_34")
INFO:tensorflow:Assets written to: /content/drive/MyDrive/graduation project (ML)/model_5/Train 11/model_34/assets
In [ ]:
# siamese_net = keras.models.load_model(r"/content/drive/MyDrive/graduation project (ML)/model_5/Train 11/model_34")
In [ ]:
print(siamese_net.summary())
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_2 (InputLayer) [(None, 150, 300, 1) 0
__________________________________________________________________________________________________
input_3 (InputLayer) [(None, 150, 300, 1) 0
__________________________________________________________________________________________________
sequential (Sequential) (None, 69120) 18315712 input_2[0][0]
input_3[0][0]
__________________________________________________________________________________________________
lambda (Lambda) (None, 69120) 0 sequential[0][0]
sequential[1][0]
__________________________________________________________________________________________________
dense (Dense) (None, 1) 69121 lambda[0][0]
==================================================================================================
Total params: 18,384,833
Trainable params: 18,155,777
Non-trainable params: 229,056
__________________________________________________________________________________________________
None
In [ ]:
img_h, img_w, img_ch = 150, 300, 1
image_shape = (img_h, img_w, img_ch)
image_shape
Out[ ]:
(150, 300, 1)
In [ ]:
path = "/content/drive/MyDrive/graduation project (ML)/datasets/1_test"
In [ ]:
dir_path, dirnames, filenames = next(os.walk(path))
filenames.sort()
len(filenames), dir_path, dirnames, filenames
Out[ ]:
(4, '/content/drive/MyDrive/graduation project (ML)/datasets/1_test', [], ['g (1).png', 'g (2).png', 'g (3).png', 'img.png'])
In [ ]:
images = []
In [ ]:
images = [path+'/'+x for x in filenames]
In [ ]:
images
Out[ ]:
['/content/drive/MyDrive/graduation project (ML)/datasets/1_test/g (1).png', '/content/drive/MyDrive/graduation project (ML)/datasets/1_test/g (2).png', '/content/drive/MyDrive/graduation project (ML)/datasets/1_test/g (3).png', '/content/drive/MyDrive/graduation project (ML)/datasets/1_test/img.png']
In [ ]:
img1 = cv2.imread(images[0],0)
img2 = cv2.imread(images[1],0)
img3 = cv2.imread(images[2],0)
img4 = cv2.imread(images[3],0)
img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img3 = cv2.resize(img3, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img4 = cv2.resize(img4, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
plt.figure(figsize = (15, 10))
plt.subplot(141)
plt.imshow(img1,'gray')
plt.title(filenames[0])
plt.subplot(142)
plt.imshow(img2,'gray')
plt.title(filenames[1])
plt.subplot(143)
plt.imshow(img3,'gray')
plt.title(filenames[2])
plt.subplot(144)
plt.imshow(img4,'gray')
plt.title(filenames[3])
plt.tight_layout()
plt.show()
In [ ]:
# list(itertools.combinations(images[:-1], 2))
In [ ]:
cv2.__version__
Out[ ]:
'4.1.2'
In [ ]:
kernel = np.ones((9,9),np.uint8) # default
def preprocessor_img(path, image_shape):
image = cv2.imread(path,0)
blured = cv2.GaussianBlur(image, (9,9), 0)
threshold, binary = cv2.threshold(blured, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel, iterations=30)
contours, hierarchies = cv2.findContours(closing, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
the_biggest_contour_by_area = max(contours, key=cv2.contourArea)
x,y,w,h = cv2.boundingRect(the_biggest_contour_by_area)
cropped = image[y:y+h, x:x+w]
resized = cv2.resize(cropped, image_shape, interpolation=cv2.INTER_LANCZOS4)
# resized_blured = cv2.GaussianBlur(resized, (9,9), 0)
threshold, resized_binary = cv2.threshold(resized, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
return resized_binary
In [ ]:
def make_pairs(images):
all_pairs = []
test_image= images[-1]
originals_images = images[:]
for i,img in enumerate(originals_images):
x1 = img
x2 = test_image
all_pairs += [[x1, x2]]
# all_pairs = list(itertools.combinations(images[:-1], 2))
pairs=[]
for ix, pair in enumerate(all_pairs):
img1 = preprocessor_img(pair[0], (img_w, img_h))
img2 = preprocessor_img(pair[1], (img_w, img_h))
# img1 = cv2.imread(pair[0],0)
# img2 = cv2.imread(pair[1],0)
# img1 = cv2.resize(img1, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
# img2 = cv2.resize(img2, (img_w, img_h), interpolation=cv2.INTER_LANCZOS4)
img1 = img1.astype('float32')
img2 = img2.astype('float32')
img1 /= 255
img2 /= 255
img1 = np.atleast_3d(img1)
img2 = np.atleast_3d(img2)
pairs.append([img1, img2])
pairs = np.array(pairs)
return pairs
In [ ]:
paris_images = make_pairs(images)
In [ ]:
paris_images.shape
Out[ ]:
(4, 2, 150, 300, 1)
In [ ]:
paris_images[:,0].shape
Out[ ]:
(4, 150, 300, 1)
In [ ]:
paris_images[0,0].shape
Out[ ]:
(150, 300, 1)
In [ ]:
for ix, pair in enumerate(paris_images):
plt.figure(figsize = (15, 10))
plt.subplot(121)
plt.imshow(pair[0],'gray')
plt.title("image_1")
plt.subplot(122)
plt.imshow(pair[1],'gray')
plt.title("image_2")
plt.tight_layout()
plt.show()
In [ ]:
# i = 0
# plt.figure(figsize = (15, 10))
# plt.subplot(121)
# plt.imshow(paris_images[i,0],'gray')
# plt.subplot(122)
# plt.imshow(paris_images[i,1],'gray')
# plt.tight_layout()
# plt.show()
In [ ]:
prediction_prob = siamese_net.predict([paris_images[:,0], paris_images[:,1]])
prediction_prob
Out[ ]:
array([[0.12951526],
[0.32044363],
[0.03803823],
[0.8471476 ]], dtype=float32)
In [ ]:
(prediction_prob.round(2)*100).flatten().astype('int')
Out[ ]:
array([13, 32, 4, 85])
In [ ]:
i = 0
prediction_prob = siamese_net.predict([np.expand_dims(paris_images[i,0],0), np.expand_dims(paris_images[i,1],0)])
prediction_prob
Out[ ]:
array([[0.12951542]], dtype=float32)
In [ ]: