-
Notifications
You must be signed in to change notification settings - Fork 0
/
hog.py
133 lines (124 loc) · 6.28 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
import numpy
from skimage.io import imread
from scipy.ndimage import convolve
from skimage import img_as_float
import math
from skimage.transform import resize
from skimage.feature import hog
HOG_CELL_ROWS = 8
HOG_CELL_COLS = 8
HOG_BIN_COUNT = 8
HOG_BLOCK_ROW_CELLS = 2
HOG_BLOCK_COL_CELLS = 2
EPSILON = 0.0000000001
def normalize_angle(angle):
while angle >= 2*math.pi:
angle -= 2*math.pi
while angle < 0:
angle += 2*math.pi
return angle
def extract_hog(img, roi, flip_left_to_right=False):
img = img.copy()[roi[1]:roi[3], roi[0]:roi[2]]
img = 0.299*img[:, :, 0] + 0.587*img[:, :, 1] + 0.114*img[:, :, 2]
img = numpy.resize(img, (128, 64))
if flip_left_to_right:
img = numpy.fliplr(img)
return hog(img, orientations=HOG_BIN_COUNT, cells_per_block=(HOG_BLOCK_ROW_CELLS, HOG_BLOCK_COL_CELLS))
# def extract_hog(img, roi, flip_left_to_right=False):
# img = img.copy()[roi[1]:roi[3], roi[0]:roi[2]]
# img = resize(img, (128, 64))
# img = img_as_float(img)
#
# if flip_left_to_right:
# img = numpy.fliplr(img)
#
# height = img.shape[0]
# width = img.shape[1]
# intensity_img = 0.299*img[:, :, 0] + 0.587*img[:, :, 1] + 0.114*img[:, :, 2]
# grad_x = numpy.empty(intensity_img.shape[:2], dtype='float64')
# grad_y = numpy.empty_like(grad_x)
# grad_power = numpy.empty_like(grad_x)
# grad_direction = numpy.empty_like(grad_x)
# x_kernel = numpy.zeros((3, 3))
# y_kernel = numpy.zeros((3, 3))
# x_kernel = [[-1, 0, 1], [-2, 0, 2], [-1, 0, 1]]
# y_kernel = [[1, 2, 1], [0, 0, 0], [-1, -2, -1]]
# grad_x = convolve(intensity_img, x_kernel)
# grad_y = convolve(intensity_img, y_kernel)
# grad_power = (grad_x**2 + grad_y**2)**0.5
# grad_direction = numpy.vectorize(math.atan2)(grad_y, grad_x)
# grad_direction = numpy.vectorize(normalize_angle)(grad_direction)
# grad_direction /= 2*math.pi
# grad_direction *= HOG_BIN_COUNT
# grad_direction = numpy.vectorize(round)(grad_direction) % HOG_BIN_COUNT
#
# # разделение на ячейки:
# cell_histograms = numpy.zeros((height // HOG_CELL_ROWS, width // HOG_CELL_COLS, HOG_BIN_COUNT), dtype='float64')
# for y in range(0, height - height % HOG_CELL_ROWS, HOG_CELL_ROWS):
# for x in range(0, width - width % HOG_CELL_COLS, HOG_CELL_COLS):
# for cell_y in range(HOG_CELL_ROWS):
# for cell_x in range(HOG_CELL_COLS):
# cell_histograms[y // HOG_CELL_ROWS, x // HOG_CELL_COLS, grad_direction[y + cell_y, x + cell_x]] += grad_power[y + cell_y, x + cell_x]
# # разделение на блоки:
# image_descriptor = numpy.empty(0)
# for y in range(height//HOG_CELL_ROWS + 1 - HOG_BLOCK_ROW_CELLS):
# for x in range(width//HOG_CELL_COLS + 1 - HOG_BLOCK_COL_CELLS):
# block_descriptor = numpy.empty(0)
# for block_y in range(HOG_BLOCK_ROW_CELLS):
# for block_x in range(HOG_BLOCK_COL_CELLS):
# block_descriptor = numpy.hstack((block_descriptor, cell_histograms[y + block_y, x + block_x]))
# block_descriptor /= math.sqrt(sum(block_descriptor ** 2) + EPSILON)
# image_descriptor = numpy.hstack((image_descriptor, block_descriptor))
# return image_descriptor
def get_hog_blocks(img):
img = 0.299*img[:, :, 0] + 0.587*img[:, :, 1] + 0.114*img[:, :, 2]
hog_vector = hog(img, orientations=HOG_BIN_COUNT, cells_per_block=(HOG_BLOCK_ROW_CELLS, HOG_BLOCK_COL_CELLS))
hog_vector = hog_vector.reshape((img.shape[0]//HOG_CELL_ROWS + 1 - HOG_BLOCK_ROW_CELLS,
img.shape[1]//HOG_CELL_COLS + 1 - HOG_BLOCK_COL_CELLS,
HOG_BIN_COUNT * HOG_BLOCK_ROW_CELLS * HOG_BLOCK_COL_CELLS))
return hog_vector
# def get_hog_blocks(img):
# img = img.copy()
# img = img_as_float(img)
#
# height = img.shape[0]
# width = img.shape[1]
# intensity_img = 0.299*img[:, :, 0] + 0.587*img[:, :, 1] + 0.114*img[:, :, 2]
# grad_x = numpy.empty(intensity_img.shape[:2], dtype='float64')
# grad_y = numpy.empty_like(grad_x)
# grad_power = numpy.empty_like(grad_x)
# grad_direction = numpy.empty_like(grad_x)
# x_kernel = numpy.zeros((3, 3))
# y_kernel = numpy.zeros((3, 3))
# x_kernel = [[-1, 0, 1], [-2, 0, 2], [-1, 0, 1]]
# y_kernel = [[1, 2, 1], [0, 0, 0], [-1, -2, -1]]
# grad_x = convolve(intensity_img, x_kernel)
# grad_y = convolve(intensity_img, y_kernel)
# grad_power = (grad_x**2 + grad_y**2)**0.5
# grad_direction = numpy.vectorize(math.atan2)(grad_y, grad_x)
# grad_direction = numpy.vectorize(normalize_angle)(grad_direction)
# grad_direction /= 2*math.pi
# grad_direction *= HOG_BIN_COUNT
# grad_direction = numpy.vectorize(round)(grad_direction) % HOG_BIN_COUNT
#
# # разделение на ячейки:
# cell_histograms = numpy.zeros((height // HOG_CELL_ROWS, width // HOG_CELL_COLS, HOG_BIN_COUNT), dtype='float64')
# for y in range(0, height - height % HOG_CELL_ROWS, HOG_CELL_ROWS):
# for x in range(0, width - width % HOG_CELL_COLS, HOG_CELL_COLS):
# for cell_y in range(HOG_CELL_ROWS):
# for cell_x in range(HOG_CELL_COLS):
# cell_histograms[y // HOG_CELL_ROWS, x // HOG_CELL_COLS, grad_direction[y + cell_y, x + cell_x]] += grad_power[y + cell_y, x + cell_x]
# # разделение на блоки:
# block_feature_number = HOG_BIN_COUNT * HOG_BLOCK_ROW_CELLS * HOG_BLOCK_COL_CELLS
# block_matrix = numpy.empty((height // HOG_CELL_ROWS + 1 - HOG_BLOCK_ROW_CELLS,
# width // HOG_CELL_COLS + 1 - HOG_BLOCK_COL_CELLS,
# block_feature_number), dtype='float64')
# for y in range(height//HOG_CELL_ROWS + 1 - HOG_BLOCK_ROW_CELLS):
# for x in range(width//HOG_CELL_COLS + 1 - HOG_BLOCK_COL_CELLS):
# block_descriptor = numpy.empty(0)
# for block_y in range(HOG_BLOCK_ROW_CELLS):
# for block_x in range(HOG_BLOCK_COL_CELLS):
# block_descriptor = numpy.hstack((block_descriptor, cell_histograms[y + block_y, x + block_x]))
# block_descriptor /= math.sqrt(sum(block_descriptor ** 2) + EPSILON)
# block_matrix[y, x] = block_descriptor
# return block_matrix