OpenCV中feature2D学习SIFT和SURF算子实现特征点提取与匹配

概述 之前的文章SURF和SIFT算子实现特征点检测简单地讲了利用SIFT和SURF算子检测特征点，在检测的基础上可以使用SIFT和SURF算子对特征点进行特征提取并使用匹配函数进行特征点的匹配。具体实现是首先采用SurfFeatureDetector检测特征点，再使用SurfDescripto

概述

      之前的文章SURF和SIFT算子实现特征点检测简单地讲了利用SIFT和SURF算子检测特征点，在检测的基础上可以使用SIFT和SURF算子对特征点进行特征提取并使用匹配函数进行特征点的匹配。具体实现是首先采用SurfFeatureDetector检测特征点，再使用SurfDescriptorExtractor计算特征点的特征向量，最后采用BruteForceMatcher暴力匹配法或者FlannBasedMatcher选择性匹配法（二者的不同）来进行特征点匹配。

      实验所用环境是opencv2.4.0+vs2008+win7，需要注意opencv2.4.X版本中SurfFeatureDetector是包含在opencv2/nonfree/features2d.hpp中，BruteForceMatcher是包含在opencv2/legacy/legacy.hpp中，FlannBasedMatcher是包含在opencv2/features2d/features2d.hpp中。

BruteForce匹配法

首先使用BruteForceMatcher暴力匹配法，代码如下：

/**
* @采用SURF算子检测特征点，对特征点进行特征提取，并使用BruteForce匹配法进行特征点的匹配
* @SurfFeatureDetector + SurfDescriptorExtractor + BruteForceMatcher
* @author holybin
*/

#include <stdio.h>
#include <iostream>
#include "opencv2/core/core.hpp"
#include "opencv2/nonfree/features2d.hpp"	//SurfFeatureDetector实际在该头文件中
#include "opencv2/legacy/legacy.hpp"	//BruteForceMatcher实际在该头文件中
//#include "opencv2/features2d/features2d.hpp"	//FlannBasedMatcher实际在该头文件中
#include "opencv2/highgui/highgui.hpp"
using namespace cv;
using namespace std;

int main( int argc, char** argv )
{
	Mat src_1 = imread( "D:\\opencv_pic\\cat3d120.jpg", CV_LOAD_IMAGE_GRAYSCALE );
	Mat src_2 = imread( "D:\\opencv_pic\\cat0.jpg", CV_LOAD_IMAGE_GRAYSCALE );

	if( !src_1.data || !src_2.data )
	{ 
		cout keypoints_1, keypoints_2;
	detector.detect( src_1, keypoints_1 );
	detector.detect( src_2, keypoints_2 );
	cout > matcher;
	vector matches;
	matcher.match( descriptors_1, descriptors_2, matches );
	cout<br>

<p>实验结果：</p>
<img  src="/inc/test.jsp?url=http%3A%2F%2Fimg.blog.csdn.net%2F20141115151204375%3Fwatermark%2F2%2Ftext%2FaHR0cDovL2Jsb2cuY3Nkbi5uZXQvaG9seWJpbg%3D%3D%2Ffont%2F5a6L5L2T%2Ffontsize%2F400%2Ffill%2FI0JBQkFCMA%3D%3D%2Fdissolve%2F70%2Fgravity%2FSouthEast&refer=http%3A%2F%2Fblog.csdn.net%2Fu012564690%2Farticle%2Fdetails%2F17370511" alt="OpenCV中feature2D学习SIFT和SURF算子实现特征点提取与匹配" ><br>
<p><span><br>
</span></p>

<h1><span>FLANN匹配法</span></h1>

<p>使用暴力匹配的结果不怎么好，下面使用FlannBasedMatcher进行特征匹配，只保留好的特征匹配点，代码如下：</p>

<pre class="brush:php;toolbar:false">/**
* @采用SURF算子检测特征点，对特征点进行特征提取，并使用FLANN匹配法进行特征点的匹配
* @SurfFeatureDetector + SurfDescriptorExtractor + FlannBasedMatcher
* @author holybin
*/

#include <stdio.h>
#include <iostream>
#include "opencv2/core/core.hpp"
#include "opencv2/nonfree/features2d.hpp"	//SurfFeatureDetector实际在该头文件中
//#include "opencv2/legacy/legacy.hpp"	//BruteForceMatcher实际在该头文件中
#include "opencv2/features2d/features2d.hpp"	//FlannBasedMatcher实际在该头文件中
#include "opencv2/highgui/highgui.hpp"
using namespace cv;
using namespace std;

int main( int argc, char** argv )
{
	Mat src_1 = imread( "D:\\opencv_pic\\cat3d120.jpg", CV_LOAD_IMAGE_GRAYSCALE );
	Mat src_2 = imread( "D:\\opencv_pic\\cat0.jpg", CV_LOAD_IMAGE_GRAYSCALE );

	if( !src_1.data || !src_2.data )
	{ 
		cout keypoints_1, keypoints_2;
	detector.detect( src_1, keypoints_1 );
	detector.detect( src_2, keypoints_2 );
	cout allMatches;
	matcher.match( descriptors_1, descriptors_2, allMatches );
	cout maxDist )
			maxDist = dist;
	}
	printf("	max dist : %f \n", maxDist );
	printf("	min dist : %f \n", minDist );

	//-- 过滤匹配点，保留好的匹配点（这里采用的标准：distance goodMatches;
	for( int i = 0; i (), 
		DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS //不显示未匹配的点
		); 
	imshow("matching result", matchImg );
	//-- 输出匹配点的对应关系
	for( int i = 0; i <br>

<p>实验结果：</p>
<img  src="/inc/test.jsp?url=http%3A%2F%2Fimg.blog.csdn.net%2F20141115151359125%3Fwatermark%2F2%2Ftext%2FaHR0cDovL2Jsb2cuY3Nkbi5uZXQvaG9seWJpbg%3D%3D%2Ffont%2F5a6L5L2T%2Ffontsize%2F400%2Ffill%2FI0JBQkFCMA%3D%3D%2Fdissolve%2F70%2Fgravity%2FSouthEast&refer=http%3A%2F%2Fblog.csdn.net%2Fu012564690%2Farticle%2Fdetails%2F17370511" alt="OpenCV中feature2D学习SIFT和SURF算子实现特征点提取与匹配" ><br>
<p><br>
</p>

<p>从第二个实验结果可以看出，经过过滤之后特征点数目从49减少到33，匹配的准确度有所上升。当然也可以使用SIFT算子进行上述两种匹配实验，只需要将SurfFeatureDetector换成SiftFeatureDetector，将SurfDescriptorExtractor换成SiftDescriptorExtractor即可。</p>
<p><br>
</p>
<h1><span>拓展</span></h1>

<p>       在FLANN匹配法的基础上，还可以进一步利用透视变换和空间映射找出已知物体（目标检测），具体来说就是利用findHomography函数利用匹配的关键点找出相应的变换，再利用perspectiveTransform函数映射点群。具体可以参考这篇文章：OpenCV中feature2D学习——SIFT和SURF算法实现目标检测。</p>
<p><br>
</p>


</iostream></stdio.h>