opencv3/C++ PHash算法圖像檢索詳解

2020-01-26 13:22:51

字體：大中小

來源：轉載

供稿：網友

PHash算法即感知哈希算法/Perceptual Hash algorithm，計算基于低頻的均值哈希．對每張圖像生成一個指紋字符串，通過對該字符串比較可以判斷圖像間的相似度．

PHash算法原理

將圖像轉為灰度圖，然后將圖片大小調整為32*32像素并通過DCT變換，取左上角的8*8像素區域。然后計算這64個像素的灰度值的均值。將每個像素的灰度值與均值對比，大于均值記為1，小于均值記為0，得到64位哈希值。

PHash算法實現

將圖片轉為灰度值

將圖片尺寸縮小為32*32

resize(src, src, Size(32, 32));

DCT變換

 Mat srcDCT;  dct(src, srcDCT);

計算DCT左上角8*8像素區域均值，求hash值

 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

  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，計算兩幅圖哈希值之間的漢明距離，漢明距離越大，兩圖片越不相似。

OpenCV實現

如圖在下圖中對比各個圖像與圖person.jpg的漢明距離，以此衡量兩圖之間的額相似度。

#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;}

輸出漢明距離：