opencv基于颜色直方图的图像比较

opencv基于颜色直方图的图像比较

图像预处理
把图像分割，把要检测的图像分割为9个小图像。比如下图为待检索的图像，将其划分为9个小部分，分别对每部分比较最后计算平均值。

计算直方图并归一化
将图像划分为9个小部分后将图像转换为HSV空间
HSV空间的图像检索使用H（色调）和S（饱和度）分量计算直方图并比较。opencv中计算直方图调用函数calcHist，将直方图归一化要调用函数normalize。
比较相似度
在得到归一化的直方图后，要度量图像的相似度。常用的相似度度量方式有5种，分别是相关性比较（Correlation）、卡方比较（Chi-Square）、十字交叉比（Intersection）、巴氏距离（Bhattacharyya）和EMD距离。
程序选择了相关性比较和EMD距离来比较图像的相关性。

程序运行环境vs2015+opencv2.4.1

#include <opencv2/core/core.hpp>    
#include "opencv2/opencv.hpp"
#include <opencv2/highgui/highgui.hpp>    
#include <iostream>  
#include <opencv2/imgproc/imgproc.hpp>  
#include <vector>  
#include <string>
#include <stdlib.h>
#include <io.h>
#include < fstream>
#include <algorithm>
//运行环境vs2015+opencv2.4.1
using namespace std;
using namespace cv;

//读入文件夹下指定格式的所有文件，本程序是读取文件夹下所有jpg图片
void getFiles(string path, string file_format, vector<string>& files)
{
	intptr_t   hFile = 0;//intptr_t和uintptr_t是什么类型:typedef long int/ typedef unsigned long int
	struct _finddata_t fileinfo;
	string p, file_formatName;
	if (0 != strcmp(file_format.c_str(), ""))
	{
		file_formatName = "\\*." + file_format;
	}
	else
	{
		file_formatName = "\\*";
	}
	if ((hFile = _findfirst(p.assign(path).append(file_formatName).c_str(), &fileinfo)) != -1)//assign方法可以理解为先将原字符串清空，然后赋予新的值作替换。
	{
		do
		{
			files.push_back(p.assign(path).append("\\").append(fileinfo.name));
		} while (_findnext(hFile, &fileinfo) == 0);
		_findclose(hFile);
	}
}
//用于数组降序排列
bool hightolow(float a, float b)
{
	return a>b;
}
//图像分割
void rectImage(Mat* rectImg, Mat oImg) {
	int m = oImg.cols;//宽
	int n = oImg.rows;//高
	rectImg[1] = oImg(Rect(0, 0, m / 3, n / 3));
	rectImg[2] = oImg(Rect(m / 3, 0, m / 3, n / 3));
	rectImg[3] = oImg(Rect(m / 3 * 2, 0, m / 3, n / 3));
	rectImg[4] = oImg(Rect(0, n / 3, m / 3, n / 3));
	rectImg[5] = oImg(Rect(m / 3, n / 3, m / 3, n / 3));
	rectImg[6] = oImg(Rect(m / 3 * 2, n / 3, m / 3, n / 3));
	rectImg[7] = oImg(Rect(0, n / 3 * 2, m / 3, n / 3));
	rectImg[8] = oImg(Rect(m / 3, n / 3 * 2, m / 3, n / 3));
	rectImg[9] = oImg(Rect(m / 3 * 2, n / 3 * 2, m / 3, n / 3));
}
//图像检索
void compareFacesByHist(string filePath, Mat orgImg)
{
	vector<Mat> src;
	vector<string> files;
	string file_format = "jpg";
	getFiles(filePath, file_format, files);
	int number = files.size();//文件数量
	//计算直方图所需参数
	vector<Mat> signature1(10);
	vector<Mat> signature2(10);
	MatND hist1, hist2;
	int scale = 10, histSize[] = { 8,8 }, ch[] = { 0,1 };
	float h_ranges[] = { 0,180 };
	float s_ranges[] = { 0,255 };
	const float* ranges[] = { h_ranges,s_ranges };
	Mat oimage[10];
	Mat image[10];
	int m = orgImg.cols;//宽
	int n = orgImg.rows;//高
	std::map<string, float> mymap;
	std::map<string, float> mymapEMD;
	//存放检索结果 大小与图像库图片数量相同
	float compare[30];
	float compareEMD[30];
	//读取图像库
	for (int i = 0; i < number; i++)
	{
		Mat img = imread(files[i]);
		src.push_back(img);
	}
	//待检索图像分割并转换为HSV空间
	rectImage(oimage, orgImg);
	for (int i = 1; i < 10; i++) {
		cvtColor(oimage[i], oimage[i], COLOR_RGB2HSV);
		
	}
	//从图像库依次读取图像检索
	for (int i = 0; i < number; i++) {
		double s = 0;
		double similarityValue = 0;
		float emd = 0;
		float semd = 0;
		mymap[files[i]] = 0;
		/*
		imge[1] = src[i](Rect(0, 0, a / 3, b / 3));
		imge[2] = src[i](Rect(a / 3, 0, a / 3, b / 3));
		imge[3] = src[i](Rect(a / 3 * 2, 0, a / 3, b / 3));
		imge[4] = src[i](Rect(0, b / 3, a / 3, b / 3));
		imge[5] = src[i](Rect(a / 3, b / 3, a / 3, b / 3));
		imge[6] = src[i](Rect(a / 3 * 2, b / 3, a / 3, b / 3));
		imge[7] = src[i](Rect(0, b / 3 * 2, a / 3, b / 3));
		imge[8] = src[i](Rect(a / 3, b / 3 * 2, a / 3, b / 3));
		imge[9] = src[i](Rect(a / 3 * 2, b / 3 * 2, a / 3, b / 3));*/
		rectImage(image, src[i]);
		for (int k = 1; k < 10; k++) {
			//计算直方图并归一化
			vector<Vec3f> sigv1, sigv2;
			calcHist(&oimage[k], 1, ch, Mat(), hist1, 2, histSize, ranges, true, false);
			normalize(hist1, hist1, 0, 1, NORM_MINMAX, -1, Mat());
			cvtColor(image[k], image[k], COLOR_RGB2HSV);
			calcHist(&image[k], 1, ch, Mat(), hist2, 2, histSize, ranges, true, false);
			normalize(hist2, hist2, 0, 1, NORM_MINMAX, -1, Mat());
			//直方图比较 改变最后一个参数可改变比较方式，当前参数为相关性比较
			similarityValue = compareHist(hist1, hist2, CV_COMP_CORREL);
			s = s + similarityValue;
			//EMD距离比较
			for (int h = 0; h < histSize[0]; h++)
				for (int s = 0; s < histSize[1]; s++) {
					float hval = hist1.at<float>(h, s);
					float hva2 = hist2.at<float>(h, s);
					if (hval != 0)
						sigv1.push_back(Vec3f(hval, (float)h, (float)s));
					sigv2.push_back(Vec3f(hva2, (float)h, (float)s));
				}

			signature1[k] = Mat(sigv1).clone().reshape(1);
			signature2[k] = Mat(sigv2).clone().reshape(1);
			emd = EMD(signature1[k], signature2[k], CV_DIST_L2);
			semd = semd + emd;
		}
		semd = semd / 9;
		s = s / 9;
		compare[i] = s;
		compareEMD[i] = semd;
		mymap[files[i]] = s;
		mymapEMD[files[i]] = semd;
		//cout << "相似度：" << s << endl;
		//cout << "emd：" << semd << endl;
	}
	sort(compare, compare + 30, hightolow);
	sort(compareEMD, compareEMD + 30);
	//输出检索出的图像 相关性比较
	/*
	for (int l = 0; l < 10; l++) {
		string path;
		for (std::map<string, float>::iterator it = mymap.begin(); it != mymap.end(); it++)
		{
			if (it->second == compare[l])
				path = it->first;
		}
		Mat image = imread(path);
		imshow("相关性" + std::to_string(compare[l]), image);
		waitKey(200);
	}*/
	//输出检索出的图像 EMD距离
	
	for (int l = 0; l < 10; l++) {
	string path;
		for (std::map<string, float>::iterator it = mymapEMD.begin(); it != mymapEMD.end(); it++)
		{
			if (it->second == compareEMD[l])
				path = it->first;
		}
		Mat image = imread(path);
		imshow("EMD" + std::to_string(compareEMD[l]), image);
		waitKey(200);
	}
	waitKey(0);

}

int main()
{
	vector<Mat> src;
	//图片库路径
	string filePath = "D:\\video-img\\testcompare\\images"; 
    //待检索图像
	Mat orgImg = imread("D:\\video-img\\testcompare\\images\\image15.jpg");
	compareFacesByHist(filePath, orgImg);
	waitKey(0);
	return 0;
}