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
|
from argparse import ArgumentParser
from collections import namedtuple
from typing import Tuple
import cv2
import matplotlib.pyplot as plt
import numpy as np
HlsSimilarity = namedtuple("HlsSimilarity", ["H", "L", "S"])
def read_image_as_hls(path: str, blur_ksize: int = 0) -> np.ndarray:
"""读取图像并转换成 HLS 格式.
由于对比的时候过于灵敏, 所以有必要加一定程度的模糊.
Returns:
image: 形状为 (H, W, 3)
"""
img = cv2.imread(path, cv2.IMREAD_COLOR)
if blur_ksize > 0:
img = cv2.blur(img, (blur_ksize, blur_ksize))
img = cv2.cvtColor(img, cv2.COLOR_BGR2HLS_FULL)
return img
def calc_image_hist_delta(img1: str, img2: str) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
"""计算两张直方图差异
Returns:
delta: (H, L, S) 三个分量的直方图差异
"""
img1 = read_image_as_hls(img1)
img2 = read_image_as_hls(img2)
hist_delta = []
for i in range(3):
h1 = cv2.calcHist([img1], [i], None, [256], [0, 255]).flatten()
h2 = cv2.calcHist([img2], [i], None, [256], [0, 255]).flatten()
hist_delta.append(h1 - h2)
return tuple(hist_delta)
def get_colors(cname: str = "hsv", size: int = 256):
"""获得连续的调色板颜色序列"""
cmap = plt.get_cmap(cname, size)
return [cmap(i)[:3] for i in range(size)]
def calc_image_similarity(img1: str, img2: str, blur_ksize: int = 15) -> HlsSimilarity:
"""计算两张图像的相似度, 具体算法见 https://docs.opencv.org/4.5.5/d8/dc8/tutorial_histogram_comparison.html
此处使用 Bhattacharyya distance, 并且加了一定量的模糊处理.
Returns:
similarities: (H, L, S) 三个分量的相似度
"""
img1 = read_image_as_hls(img1, blur_ksize)
img2 = read_image_as_hls(img2, blur_ksize)
similarities = [0, 0, 0]
similarities[0] = 1 - cv2.compareHist(
cv2.calcHist([img1], [0], None, [256], [0, 255]), cv2.calcHist([img2], [0], None, [256], [0, 255]), 3
)
similarities[1] = 1 - cv2.compareHist(
cv2.calcHist([img1], [1], None, [256], [0, 255]), cv2.calcHist([img2], [1], None, [256], [0, 255]), 3
)
similarities[2] = 1 - cv2.compareHist(
cv2.calcHist([img1], [2], None, [256], [0, 255]), cv2.calcHist([img2], [2], None, [256], [0, 255]), 3
)
return HlsSimilarity(*similarities)
def draw_image_hist_delta(img1: str, img2: str, path: str, title: str = "Hist Delta of HLS Image (First - Second)"):
"""绘制差异直方图情况"""
delta_H, delta_L, delta_S = calc_image_hist_delta(img1, img2)
fig, axes = plt.subplots(3, 1, squeeze=True, figsize=(12, 16), dpi=300)
fig.subplots_adjust(left=0.1, right=0.9, top=0.93, bottom=0.07)
fig.suptitle(title)
fig.supxlabel("Pixel Value")
fig.supylabel("Pixel Count")
axes: list[plt.Axes]
ax_H, ax_L, ax_S = axes
x_len = len(delta_H)
ax_H.bar(np.arange(x_len), delta_H, color=get_colors("hsv", x_len))
ax_H.set_title("Delta of H Channel")
ax_H.set_xticks(np.arange(0, x_len, 10), np.arange(0, x_len, 10), rotation=-45)
x_len = len(delta_L)
ax_L.bar(np.arange(x_len), delta_L, color=get_colors("gray", x_len))
ax_L.set_title("Delta of L Channel")
ax_L.set_xticks(np.arange(0, x_len, 10), np.arange(0, x_len, 10), rotation=-45)
x_len = len(delta_S)
ax_S.bar(np.arange(x_len), delta_S, color=get_colors("gray", x_len))
ax_S.set_title("Delta of S Channel")
ax_S.set_xticks(np.arange(0, x_len, 10), np.arange(0, x_len, 10), rotation=-45)
fig.savefig(path)
def make_diff_image(img1: str, img2: str, path: str) -> bool:
"""叠图查看差异, 如果尺寸不一样或者保存失败返回 False."""
img1 = read_image_as_hls(img1)
img2 = read_image_as_hls(img2)
if img1.shape != img2.shape:
return False
img1_H, img1_L, img1_S = cv2.split(img1)
img2_H, img2_L, img2_S = cv2.split(img2)
diff_H = np.min([img1_H - img2_H, img2_H - img1_H], axis=0)
diff_H = cv2.normalize(diff_H, None, 0, 255, cv2.NORM_MINMAX)
diff_L = cv2.absdiff(img1_L, img2_L)
diff_S = cv2.absdiff(img1_S, img2_S)
diff_img = np.concatenate([diff_H, diff_L, diff_S], axis=0)
return cv2.imwrite(path, diff_img)
if __name__ == "__main__":
parser = ArgumentParser()
parser.add_argument("img1")
parser.add_argument("img2")
parser.add_argument("--out-hist", default="hist.png")
parser.add_argument("--out-diff", default="diff.png")
args = parser.parse_args()
sim = calc_image_similarity(args.img1, args.img2)
print(sim)
draw_image_hist_delta(args.img1, args.img2, args.out_hist)
if not make_diff_image(args.img1, args.img2, args.out_diff):
print("尺寸不同, 无法计算差分")
|
