PHash算法即感知哈希算法/Perceptual Hash algorithm，计算基于低频的均值哈希．对每张图像生成一个指纹字符串，通过对该字符串比较可以判断图像间的相似度．
PHash算法原理
将图像转为灰度图，然后将图片大小调整为32*32像素并通过DCT变换，取左上角的8*8像素区域。然后计算这64个像素的灰度值的均值。将每个像素的灰度值与均值对比，大于均值记为1，小于均值记为0，得到64位哈希值。
PHash算法实现
将图片转为灰度值
将图片尺寸缩小为32*32
|
1
|
resize(src, src, Size(32, 32)); |
DCT变换
|
1
2
|
Mat srcDCT; dct(src, srcDCT); |
计算DCT左上角8*8像素区域均值，求hash值
|
1
2
3
4
5
6
7
8
9
10
|
double sum = 0;for (int i = 0; i < 8; i++) for (int j = 0; j < 8; j++) sum += srcDCT.at<float>(i,j);double average = sum/64;Mat phashcode= Mat::zeros(Size(8, 8), CV_8U);for (int i = 0; i < 8; i++) for (int j = 0; j < 8; j++) phashcode.at<char>(i,j) = srcDCT.at<float>(i,j) > average ? 1:0; |
hash值匹配
|
1
2
3
|
int d = 0;for (int n = 0; n < srchash.size[1]; n++) if (srchash.at<uchar>(0,n) != dsthash.at<uchar>(0,n)) d++; |
即，计算两幅图哈希值之间的汉明距离，汉明距离越大，两图片越不相似。
OpenCV实现
如图在下图中对比各个图像与图person.jpg的汉明距离，以此衡量两图之间的额相似度。
|
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
|
#include <iostream> #include <stdio.h>#include <fstream>#include <io.h>#include <string>#include <opencv2\opencv.hpp> #include <opencv2\core\core.hpp>#include <opencv2\core\mat.hpp>using namespace std; using namespace cv; int fingerprint(Mat src, Mat* hash);int main(){ Mat src = imread("E:\\image\\image\\image\\person.jpg", 0); if(src.empty()) { cout << "the image is not exist" << endl; return -1; } Mat srchash, dsthash; fingerprint(src, &srchash); for(int i = 1; i <= 8; i++) { string path0 = "E:\\image\\image\\image\\person"; string number; stringstream ss; ss << i; ss >> number; string path = "E:\\image\\image\\image\\person" + number +".jpg"; Mat dst = imread(path, 0); if(dst.empty()) { cout << "the image is not exist" << endl; return -1; } fingerprint(dst, &dsthash); int d = 0; for (int n = 0; n < srchash.size[1]; n++) if (srchash.at<uchar>(0,n) != dsthash.at<uchar>(0,n)) d++; cout <<"person" << i <<" distance= " <<d<<"\n"; } system("pause"); return 0;}int fingerprint(Mat src, Mat* hash){ resize(src, src, Size(32, 32)); src.convertTo(src, CV_32F); Mat srcDCT; dct(src, srcDCT); srcDCT = abs(srcDCT); double sum = 0; for (int i = 0; i < 8; i++) for (int j = 0; j < 8; j++) sum += srcDCT.at<float>(i,j); double average = sum/64; Mat phashcode= Mat::zeros(Size(8, 8), CV_8U); for (int i = 0; i < 8; i++) for (int j = 0; j < 8; j++) phashcode.at<char>(i,j) = srcDCT.at<float>(i,j) > average ? 1:0; *hash = phashcode.reshape(0,1).clone(); return 0;} |
输出汉明距离：
可以看出若将阈值设置为20则可将后三张其他图片筛选掉。
以上这篇opencv3/C++ PHash算法图像检索详解就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多支持服务器之家。
原文链接：https://blog.csdn.net/akadiao/article/details/79779634
