OpenCV 图像相似度对比算法
2024-04-12 本文已影响0人
LiuJP
图像哈希：均值哈希、感知哈希、差值哈希
1、均值哈希
long aHash(const Mat &srcMat) {
Mat dstMat;
resize(srcMat, dstMat, Size(8, 8));
// cvtColor(dstMat, dstMat, COLOR_BGR2GRAY);
double iAvg = mean(dstMat)[0];
// printf("均值哈希: %lf \n", iAvg);
int p = 1;
long value = 0;
for (int i = 0; i < 8; i++) {
for (int j = 0; j < 8; j++) {
p <<= 1;
if (dstMat.at<uchar>(i, j) >= iAvg) {
value |= p;
}
}
}
// printf("%ld ;\n", value);
dstMat.release();
return value;
}
2、余玄感知哈希算法；
/**
* 感知哈希算法
* @param img
*/
void pHash32(cv::Mat img, bool gray,unsigned char **result) {
cv::Mat dstMat;
cv::resize(img, dstMat, cv::Size(32, 32));
if (gray)
cv::cvtColor(dstMat, dstMat, cv::COLOR_BGR2GRAY);
dstMat.convertTo(dstMat, CV_32F);
cv::Mat srcDCT;
cv::dct(dstMat, srcDCT);
srcDCT = cv::abs(srcDCT);
cv::Mat finalMat;
cv::resize(srcDCT,finalMat,cv::Size(8,8));
double average = cv::mean(finalMat)[0];
// printf("感知哈希2 %lf \n", average);
for (int i = 0; i < 8; i++) {
for (int j = 0; j < 8; j++) {
if (srcDCT.at<float>(i, j) >= average) {
result[i][j] = 1;
} else {
result[i][j] = 0;
}
}
}
// printf("%ld ;\n", value);
dstMat.release();
srcDCT.release();
}
延伸：三种感知算法：
# ---------------------------------------------------------------------------------------------------------------------
# 测试：为什么要缩放DCT？DCT缩放方式有哪些？不同DCT缩放方式有何不同？不进行DCT缩放效果会怎么样？
# ---------------------------------------------------------------------------------------------------------------------
import cv2
import time
import numpy as np
import matplotlib.pyplot as plt
# DCT变换后：无特征频率区域缩放，使用整个32x32图像块的频率分布，计算整个DCT系数的均值，并根据这个均值生成哈希值。
def get_pHash1(img_path):
img = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
# plt.imshow(img, cmap='gray')
# plt.show()
img = cv2.resize(img, (32, 32), cv2.INTER_CUBIC)
# plt.imshow(img, cmap='gray')
# plt.show()
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# plt.imshow(img_gray, cmap='gray')
# plt.show()
img_dct = cv2.dct(np.float32(img_gray))
# 显示DCT系数的图像
# dct_scaled = cv2.normalize(img_dct, None, 0, 255, cv2.NORM_MINMAX)
# img_dct_scaled = dct_scaled.astype(np.uint8)
# plt.imshow(img_dct_scaled, cmap='gray')
# plt.show()
img_avg = np.mean(img_dct)
# print(f"DCT变换后图像块的均值={img_avg}")
img_hash_str = get_img_hash_binary(img_dct, img_avg)
# print(f"图像的二进制哈希值={img_hash_str}")
img_hash = get_img_hash(img_hash_str)
return img_hash
# DCT变换后：将DCT系数的大小显式地调整为8x8，使用8x8的DCT系数块的频率分布，计算调整后的DCT系数的均值，并生成哈希值。
def get_pHash2(img_path):
img = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
# plt.imshow(img, cmap='gray')
# plt.show()
img = cv2.resize(img, (32, 32), cv2.INTER_CUBIC)
# plt.imshow(img, cmap='gray')
# plt.show()
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# plt.imshow(img_gray, cmap='gray')
# plt.show()
img_dct = cv2.dct(np.float32(img_gray))
img_dct.resize(8, 8)
# 显示DCT系数的图像
# dct_scaled = cv2.normalize(img_dct, None, 0, 255, cv2.NORM_MINMAX)
# img_dct_scaled = dct_scaled.astype(np.uint8)
# plt.imshow(img_dct_scaled, cmap='gray')
# plt.show()
img_avg = np.mean(img_dct)
# print(f"DCT变换后图像块的均值={img_avg}")
img_hash_str = get_img_hash_binary(img_dct, img_avg)
# print(f"图像的二进制哈希值={img_hash_str}")
img_hash = get_img_hash(img_hash_str)
return img_hash
# DCT变换后：只提取DCT系数的左上角8x8块的信息，然后计算这个块的均值。此法只考虑图像一小部分的频率分布，并生成哈希值。
def get_pHash3(img_path):
img = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
# plt.imshow(img, cmap='gray')
# plt.show()
img = cv2.resize(img, (32, 32), cv2.INTER_CUBIC)
# plt.imshow(img, cmap='gray')
# plt.show()
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# plt.imshow(img_gray, cmap='gray')
# plt.show()
img_dct = cv2.dct(np.float32(img_gray))
dct_roi = img_dct[0:8, 0:8]
# 显示DCT系数的图像
# dct_scaled = cv2.normalize(dct_roi, None, 0, 255, cv2.NORM_MINMAX)
# img_dct_scaled = dct_scaled.astype(np.uint8)
# plt.imshow(img_dct_scaled, cmap='gray')
# plt.show()
img_avg = np.mean(dct_roi)
# print(f"DCT变换后图像块的均值={img_avg}")
img_hash_str = get_img_hash_binary(dct_roi, img_avg)
# print(f"图像的二进制哈希值={img_hash_str}")
img_hash = get_img_hash(img_hash_str)
return img_hash
def get_img_hash_binary(img_dct, img_avg):
img_hash_str = ''
for i in range(8):
img_hash_str += ''.join(map(lambda i: '0' if i < img_avg else '1', img_dct[i]))
# print(f"图像的二进制哈希值={img_hash_str}")
return img_hash_str
def get_img_hash(img_hash_str):
img_hash = ''.join(map(lambda x:'%x' % int(img_hash_str[x : x + 4], 2), range(0, 64, 4)))
# print(f"图像可识别的哈希值={img_hash}")
return img_hash
# 汉明距离：计算两个等长字符串（通常是二进制字符串或位字符串）之间的汉明距离。用于确定两个等长字符串在相同位置上不同字符的数量。
def hamming_distance(s1, s2):
# 检查这两个字符串的长度是否相同。如果长度不同，它会引发 ValueError 异常，因为汉明距离只适用于等长的字符串
if len(s1) != len(s2):
raise ValueError("Input strings must have the same length")
distance = 0
for i in range(len(s1)):
# 遍历两个字符串的每个字符，比较它们在相同位置上的值。如果发现不同的字符，将 distance 的值增加 1
if s1[i] != s2[i]:
distance += 1
return distance
# ======================================== 测试场景一 ========================================
# img = 'img_test/apple-01.jpg'
# img_hash1 = get_phash1(img)
# img_hash2 = get_phash2(img)
# img_hash3 = get_phash3(img)
# print(f"方式一：DCT变换后，无DCT特征频率区域缩放，获得图像的二进制哈希值={img_hash1}")
# print(f"方式二：DCT变换后，将DCT系数显式调整为8x8，获得图像的二进制哈希值={img_hash2}")
# print(f"方式三：DCT变换后，只提取DCT系数左上角8x8像素，获得图像的二进制哈希值={img_hash3}")
# ======================================== 测试场景二 ========================================
time_start = time.time()
img_1 = 'img_test/apple-01.jpg'
img_2 = 'img_test/apple-02.jpg'
img_3 = 'img_test/apple-03.jpg'
img_4 = 'img_test/apple-04.jpg'
img_5 = 'img_test/apple-05.jpg'
img_6 = 'img_test/apple-06.jpg'
img_7 = 'img_test/apple-07.jpg'
img_8 = 'img_test/apple-08.jpg'
img_9 = 'img_test/apple-09.jpg'
img_10 = 'img_test/pear-001.jpg'
# ------------------------------------- 测试场景二：方式一 --------------------------------------
# img_hash1 = get_pHash1(img_1)
# img_hash2 = get_pHash1(img_2)
# img_hash3 = get_pHash1(img_3)
# img_hash4 = get_pHash1(img_4)
# img_hash5 = get_pHash1(img_5)
# img_hash6 = get_pHash1(img_6)
# img_hash7 = get_pHash1(img_7)
# img_hash8 = get_pHash1(img_8)
# img_hash9 = get_pHash1(img_9)
# img_hash10 = get_pHash1(img_10)
# ------------------------------------- 测试场景二：方式二 --------------------------------------
img_hash1 = get_pHash2(img_1)
img_hash2 = get_pHash2(img_2)
img_hash3 = get_pHash2(img_3)
img_hash4 = get_pHash2(img_4)
img_hash5 = get_pHash2(img_5)
img_hash6 = get_pHash2(img_6)
img_hash7 = get_pHash2(img_7)
img_hash8 = get_pHash2(img_8)
img_hash9 = get_pHash2(img_9)
img_hash10 = get_pHash2(img_10)
# ------------------------------------- 测试场景二：方式三 --------------------------------------
# img_hash1 = get_pHash3(img_1)
# img_hash2 = get_pHash3(img_2)
# img_hash3 = get_pHash3(img_3)
# img_hash4 = get_pHash3(img_4)
# img_hash5 = get_pHash3(img_5)
# img_hash6 = get_pHash3(img_6)
# img_hash7 = get_pHash3(img_7)
# img_hash8 = get_pHash3(img_8)
# img_hash9 = get_pHash3(img_9)
# img_hash10 = get_pHash3(img_10)
distance1 = hamming_distance(img_hash1, img_hash1)
distance2 = hamming_distance(img_hash1, img_hash2)
distance3 = hamming_distance(img_hash1, img_hash3)
distance4 = hamming_distance(img_hash1, img_hash4)
distance5 = hamming_distance(img_hash1, img_hash5)
distance6 = hamming_distance(img_hash1, img_hash6)
distance7 = hamming_distance(img_hash1, img_hash7)
distance8 = hamming_distance(img_hash1, img_hash8)
distance9 = hamming_distance(img_hash1, img_hash9)
distance10 = hamming_distance(img_hash1, img_hash10)
time_end = time.time()
print(f"图片名称：{img_1}，图片HASH：{img_hash1}，与图片1的近似值（汉明距离）：{distance1}")
print(f"图片名称：{img_2}，图片HASH：{img_hash2}，与图片1的近似值（汉明距离）：{distance2}")
print(f"图片名称：{img_3}，图片HASH：{img_hash3}，与图片1的近似值（汉明距离）：{distance3}")
print(f"图片名称：{img_4}，图片HASH：{img_hash4}，与图片1的近似值（汉明距离）：{distance4}")
print(f"图片名称：{img_5}，图片HASH：{img_hash5}，与图片1的近似值（汉明距离）：{distance5}")
print(f"图片名称：{img_6}，图片HASH：{img_hash6}，与图片1的近似值（汉明距离）：{distance6}")
print(f"图片名称：{img_7}，图片HASH：{img_hash7}，与图片1的近似值（汉明距离）：{distance7}")
print(f"图片名称：{img_8}，图片HASH：{img_hash8}，与图片1的近似值（汉明距离）：{distance8}")
print(f"图片名称：{img_9}，图片HASH：{img_hash9}，与图片1的近似值（汉明距离）：{distance9}")
print(f"图片名称：{img_10}，图片HASH：{img_hash10}，与图片1的近似值（汉明距离）：{distance10}")
print(f"耗时：{time_end - time_start}")
————————————————
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
原文链接：https://blog.csdn.net/itanping/article/details/134022715
3、差值哈希算法
def get_dHash(img_path):
# 读取图像：通过OpenCV的imread加载RGB图像
img_rgb = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
# 缩小图像：使用OpenCV的resize函数将图像缩放为9x8像素，采用Cubic插值方法进行图像重采样
img_resize = cv2.resize(img_rgb, (9, 8), cv2.INTER_CUBIC)
# 图像灰度化：将彩色图像转换为灰度图像
img_gray = cv2.cvtColor(img_resize, cv2.COLOR_BGR2GRAY)
# 计算差异值：获得图像二进制字符串
img_hash_str = ''
# 遍历图像的像素，比较相邻像素之间的灰度值，根据强弱增减差异情况生成一个二进制哈希值
# 外层循环，遍历图像的行（垂直方向），范围是从0到7
for i in range(8):
# 内层循环，遍历图像的列（水平方向），范围也是从0到7
for j in range(8):
# 比较当前像素 img[i, j] 与下一个像素 img[i, j + 1] 的灰度值
if img_gray[i, j] > img_gray[i, j + 1]:
# 如果当前像素的灰度值大于下一个像素的灰度值（灰度值增加），将1添加到名为 hash 的列表中
img_hash_str += '1'
else:
# 否则灰度值弱减，将0添加到名为 hash 的列表中
img_hash_str += '0'
# print(f"图像的二进制哈希值={img_hash_str}")
# 生成哈希值：生成图像可识别哈希值
img_hash = ''.join(map(lambda x:'%x' % int(img_hash_str[x : x + 4], 2), range(0, 64, 4)))
return img_hash
4、单通道直方图
def histogram(image1, image2):
# 灰度直方图算法
# 计算单通道的直方图的相似值
hist1 = cv2.calcHist([image1], [0], None, [256], [0.0, 255.0])
hist2 = cv2.calcHist([image2], [0], None, [256], [0.0, 255.0])
# 计算直方图的重合度
degree = 0
for i in range(len(hist1)):
if hist1[i] != hist2[i]:
degree = degree + \
(1 - abs(hist1[i] - hist2[i]) / max(hist1[i], hist2[i]))
else:
degree = degree + 1
degree = degree / len(hist1)
return degree
