-
Notifications
You must be signed in to change notification settings - Fork 3
/
hog.py
203 lines (173 loc) · 7.36 KB
/
hog.py
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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
import cv2
import numpy as np
import os
from sklearn import svm
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
import skimage.io as io
import os
import numpy as np
from skimage import feature
from sklearn import svm
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from PIL import Image
import cv2
from skimage.feature import local_binary_pattern
from sklearn.decomposition import PCA
# Define the directory where the hand gesture images are stored
dataset_dir = "dataset_sample\Women"
images = []
labels = []
descriptors = []
features=[]
arr=[]
# Define the HOG parameters
pixels_per_cell = (8, 8)
cells_per_block = (2, 2)
num_orientations = 9
for sub_dir in os.listdir(dataset_dir):
sub_dir_path = os.path.join(dataset_dir, sub_dir)
if not os.path.isdir(sub_dir_path):
continue
# Iterate through each image file in the subdirectory
for file_name in os.listdir(sub_dir_path):
if not file_name.endswith(".JPG"):
continue
image_path = os.path.join(sub_dir_path, file_name)
# Load the image and compute its HOG features
# image = np.asarray(Image.open(image_path))
# print(image)
# cv2.imshow("image",image)
# break
image = np.asarray(Image.open(image_path).convert("L"))
# num_channels = image.shape[-1]
# # image = cv2.cvtColor(image, cv2.COLOR_RGBA2RGB)
# # blur = cv2.GaussianBlur(image, (3,3), 0)
# # print(('im',(image)))
# # print(('blur',(blur)))
# # print((np.min(blur)))
# # print((np.max(blur)))
# # Change color-space from BGR -> HSV
# # if num_channels == 3:
# # # hsv = cv2.cvtColor(blur, cv2.COLOR_RGB2HSV)
# # hsv = cv2.cvtColor(blur, cv2.COLOR_BGR2YCR_CB)
# # else:
# # hsv = cv2.cvtColor(blur, cv2.COLOR_BGRA2YCR_CB)
# # hsv = cv2.cvtColor(blur, cv2.COLOR_RGBA2HSV) # Already grayscale
# # hsv = cv2.cvtColor(blur, cv2.COLOR_BGR2YCR_CB)
# gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# print(image.shape)
# hsv= cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
# # hsv = cv2.cvtColor(blur, cv2.COLOR_RGB2HSV)
# print(('hsv',(hsv)))
# print((np.min(hsv)))
# print((np.max(hsv)))
# image = cv2.inRange(hsv, np.array([0,40,30],dtype="uint8"), np.array([43,255,254],dtype="uint8"))
# cv2.imshow("image",image)
# print(('image',(image)))
# print((np.min(image)))
# print((np.max(image)))
# arr.append(image)
# image = cv2.resize(image, (600,400))
# sift = cv2.SIFT_create()
# surf = cv2.xfeatures2d.SURF_create(128)
# Convert the image to grayscale if it has three channels
# if num_channels == 3:
# gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# else:
# gray = image # Already grayscale
# kp, des = sift.detectAndCompute(gray, None)
# # print(kp[0].pt[0])
# if des is not None:
# mean=np.mean(des,axis=0)
# descriptors.append(mean)
# Convert the image to grayscale
# gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# # Apply adaptive thresholding to the grayscale image
# thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2)
# # Apply a median blur to the thresholded image to remove noise
# blur = cv2.medianBlur(thresh, 7)
# # Convert the image to the YCrCb color space
# ycrcb = cv2.cvtColor(image, cv2.COLOR_BGR2YCrCb)
# # Apply a skin color range filter to the YCrCb image
# lower_skin = np.array([0, 135, 85])
# upper_skin = np.array([255, 180, 135])
# mask = cv2.inRange(ycrcb, lower_skin, upper_skin)
# print(mask)
labels.append(sub_dir)
# image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
hog_features = feature.hog(image, pixels_per_cell=pixels_per_cell,
cells_per_block=cells_per_block,
orientations=num_orientations ,visualize=True)
# print(len(hog_features))
# hog_features = np.array(hog_features)
# hog_features=np.concatenate(hog_features)
# pca = PCA(n_components=50)
# print(des)
# hog_features = pca.fit_transform(hog_features.reshape(1,-1))
# # Add the HOG features and label to the lists
print(hog_features.shape)
print(hog_features)
features.append(hog_features)
# descriptors = np.vstack(descriptors)
# descriptors.append(des)
# descriptors = np.array(descriptors)
features = np.array(features)
# total=np.concatenate((descriptors, features), axis=1)
# print('hog',features)
# print('hof shape',features.shape)
# surf_des=np.array(surf_des)
labels = np.array(labels)
# print(surf_des.shape)
# descriptors = descriptors.reshape(descriptors.shape[0], descriptors.shape[1])
# descriptors = np.reshape(descriptors, (len(labels), -1))
# print('sift shape',descriptors.shape)
print(labels.shape)
# print('sift',descriptors)
# for image in images:
# kp, des = sift.detectAndCompute(image, None)
# descriptors.append(des)
# descriptors = np.array(descriptors)
# labels = np.array(labels)
# Split the dataset into training and testing sets
train_features, test_features, train_labels, test_labels = train_test_split(
features, labels, test_size=0.25, random_state=42)
print('Shape of train_images:', train_features.shape)
print('Shape of train_labels:', train_labels.shape)
print('Shape of test_images:', test_features.shape)
print('Shape of test_labels:', test_labels.shape)
# Train a Support Vector Machine (SVM) classifier
svm_classifier = svm.SVC(kernel="linear")
svm_classifier.fit(train_features, train_labels)
# Predict the labels of the test set using the trained SVM classifier
predicted_labels = svm_classifier.predict(test_features)
# Compute the accuracy of the SVM classifier
accuracy = accuracy_score(test_labels, predicted_labels)
print("Accuracy: {:.2f}%".format(accuracy * 100))
# # Split data into training and testing sets
# train_descriptors, test_descriptors, train_labels, test_labels = train_test_split(
# descriptors, labels, test_size=0.2, random_state=42)
# # Train the SVM classifier
# clf = svm.SVC(kernel='linear')
# clf.fit(train_descriptors, train_labels)
# # Load new test images
# test_images = []
# test_dir_path = 'test_images'
# for filename in os.listdir(test_dir_path):
# img = cv2.imread(os.path.join(test_dir_path, filename))
# gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# test_images.append(gray)
# # Extract SIFT features from test images
# test_descriptors = []
# for image in test_images:
# kp, des = sift.detectAndCompute(image, None)
# test_descriptors.append(des)
# test_descriptors = np.array(test_descriptors)
# # Classify test images using SVM classifier
# predicted_labels = clf.predict(test_descriptors)
# # Print predicted labels
# print(predicted_labels)
# # Evaluate accuracy on test set
# accuracy = accuracy_score(test_labels, predicted_labels)
# print("Accuracy:", accuracy)