Author: Steven Zhai
Copyright Notice: The copyright belongs to the author. For commercial reprints, please contact the author for authorization. For non-commercial reprints, please indicate the source

function prototype

void matchTemplate( InputArray image, InputArray templ,OutputArray result, int method, InputArray mask = noArray());

Parameter Description

Image of type InputArray, input image.
templ of type InputArray, the image to be matched.
A result of type OutputArray, which outputs the matching result.
Int type method, matching method, see below for details.
Mask of type InputArray, mask.

Introduction to matching methods

The template matching function has 6 matching methods. It will be found that if the original image is 1000*1000 and the template is 100*100, the resulting image size is 900*900. The formulas and principles corresponding to each method are as follows.

1.CV_TM_SQDIFF, square difference matching method. The smaller the value obtained, the better the match.

$R(x, y)=\sum(I(x+m, y+n)-T(m, n))^{2}$

2. CV_TM_SQDIFF_NORMED, normalized squared difference matching method. The smaller the value obtained, the better the match.

$R(x, y)=\frac{\sum(I(x+m, y+n)-T(m, n))^{2}}{\sqrt{\sum T(m, n)^{2} \sum I(x+m, y+n)^{2}}}$

3.CV_TM_CCORR, correlation matching method. The higher the value obtained, the better the match. This method is susceptible to background interference.

$R(x, y)=\sum(I(x+m, y+n) \cdot T(m, n))$

4. CV_TM_CCORR_NORMED, normalized correlation matching method. The higher the value obtained, the better the match. This method is susceptible to background interference.

$R(x, y)=\frac{\sum(I(x+m, y+n) \cdot T(m, n))}{\sqrt{\sum T(m, n)^{2} \sum I(x+m, y+n)^{2}}}$

5. CV_TM_CCOEFF, the correlation coefficient method, is also a zero-mean cross-correlation method. After subtracting the mean, the correlation is calculated. The higher the value obtained, the better the match. This method has better applicability.

$R(x, y)=\sum((I(x+m, y+n)-\bar{I}) \cdot(T(m, n)-\bar{T}))$

6. CV_TM_CCOEFF_NORMED, normalized correlation coefficient method. The higher the value obtained, the better the match. This method has better applicability.

$R(x, y)=\frac{\sum((I(x+m, y+n)-\bar{I}) \cdot(T(m, n)-\bar{T}))}{\sqrt{\sum(T(m, n)-T)^{2} \sum(I(x+m, y+n)-I)^{2}}}$

test code

#include <iostream>  
#include <opencv2\opencv.hpp>
#include <opencv2/highgui/highgui.hpp> 

using namespace cv;
using namespace std;

int main()
{
	cv::Mat src = imread("test1.jpg");
	cv::Mat sample = imread("t.png");
	// 匹配
	cv::Mat result;
	matchTemplate(src, sample, result, CV_TM_CCOEFF);
	// 归一化
	normalize(result, result, 0, 1, NORM_MINMAX, -1, Mat());
	// 获取最小值
	double minValue; double maxValue; Point minLocation; Point maxLocation;
	Point matchLocation;
	minMaxLoc(result, &minValue, &maxValue, &minLocation, &maxLocation, Mat());
	matchLocation = maxLocation;
	// 框选结果
	cv::Mat draw = src.clone();
	rectangle(draw, matchLocation, Point(matchLocation.x + sample.cols, matchLocation.y + sample.rows), Scalar(255, 0, 0), 2, 8, 0);
	
	imshow("src", src);
	imshow("sample", sample);
	imshow("draw", draw);
	waitKey(0);
	return 0;
}

Test effect

If the article helps you, you can give me a like and let me know, I will be very happy~ Come on!

OpenCV - template matching cv::matchTemplate

function prototype

Parameter Description

Introduction to matching methods

test code

Test effect

Guess you like