opencv——SURF特征点检测并使用Flann算法匹配

    记录一下学习过程，话不多说贴代码：

#include <opencv2/core/core.hpp>
#include <opencv2/features2d/features2d.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/flann/flann.hpp>
#include <opencv2/xfeatures2d/nonfree.hpp>
#include <opencv2/xfeatures2d.hpp>
#include <iostream>

int main() {
	cv::Mat srcImage1 = cv::imread("img1.png", 1);
	cv::Mat srcImage2 = cv::imread("img2.png", 1);
	if (!srcImage1.data || !srcImage2.data) {
		std::cout << "No images" << std::endl;
	}
	int minHessian = 400;
	cv::Ptr<cv::xfeatures2d::SurfFeatureDetector> detector = cv::xfeatures2d::SurfFeatureDetector::create(minHessian);
	
	//keypoint存储着特征点像素的坐标
	std::vector<cv::KeyPoint> keypoint1, keypoint2;
	cv::Mat dstImage1, dstImage2;
	//特征点检测，并计算特征点的特征向量
	detector->detect(srcImage1, keypoint1);
	detector->detect(srcImage2, keypoint2);
	detector->compute(srcImage1, keypoint1, dstImage1);
	detector->compute(srcImage2, keypoint2, dstImage2);
	//dstImage里面存储的是SURF特征，64*X,每个SURF特征是一个64维的向量
	std::cout << dstImage1.size() << std::endl;
	//keypImage是标注了特征点的原图像
	cv::Mat keypImage1, keypImage2;
	cv::drawKeypoints(srcImage1, keypoint1, keypImage1);
	cv::drawKeypoints(srcImage2, keypoint2, keypImage2);
	//cv::imshow("keypoint1", keypImage1);
	//cv::imshow("keypoint2", keypImage2);
	
	//特征点匹配
	cv::Ptr<cv::DescriptorMatcher> matcher = cv::DescriptorMatcher::create("FlannBased");
	std::vector<cv::DMatch> match;
	matcher->match(dstImage1, dstImage2, match);
	//根据粗匹配的结果筛选（通过比较匹配间的距离）优匹配的特征点
	double mindist = 100;
	double maxdist = 0;
	for (int i = 0; i < dstImage1.rows; i++) {
		double dist = match[i].distance;
		if (dist < mindist) mindist = dist;
		if (dist > maxdist) maxdist = dist;
	}
	std::cout << mindist << std::endl;
	std::cout << maxdist << std::endl;
	std::vector<cv::DMatch> goodmatch;
	for (int i = 0; i < dstImage1.rows; i++) {
		//距离满足的具体条件视情况而定
		if (match[i].distance == mindist) {
			goodmatch.push_back(match[i]);
			std::cout << match[i].queryIdx << "-------" << match[i].trainIdx << std::endl;
		}
	}

	cv::Mat matchImage;
	cv::drawMatches(srcImage1, keypoint1, srcImage2, keypoint2, goodmatch, matchImage);
	std::cout << goodmatch.size() << std::endl;
	cv::imshow("goodmatch", matchImage);
	cv::waitKey(0);
	return 0;
}

    实际运行的效果：

    代码中需要了解一下opencv里Dmatch数据结构和detector的数据类型：

//////////////////////////////// DMatch /////////////////////////////////

/** @brief Class for matching keypoint descriptors

query descriptor index, train descriptor index, train image index, and distance between
descriptors.
*/
class CV_EXPORTS_W_SIMPLE DMatch
{
public:
    CV_WRAP DMatch();
    CV_WRAP DMatch(int _queryIdx, int _trainIdx, float _distance);
    CV_WRAP DMatch(int _queryIdx, int _trainIdx, int _imgIdx, float _distance);

    CV_PROP_RW int queryIdx; // query descriptor index
    CV_PROP_RW int trainIdx; // train descriptor index
    CV_PROP_RW int imgIdx;   // train image index

    CV_PROP_RW float distance;

    // less is better
    bool operator<(const DMatch &m) const;
};

template<> class DataType<DMatch>
{
public:
    typedef DMatch      value_type;
    typedef int         work_type;
    typedef int         channel_type;

    enum { generic_type = 0,
           depth        = DataType<channel_type>::depth,
           channels     = (int)(sizeof(value_type)/sizeof(channel_type)), // 4
           fmt          = DataType<channel_type>::fmt + ((channels - 1) << 8),
           type         = CV_MAKETYPE(depth, channels)
         };

    typedef Vec<channel_type, channels> vec_type;
};

    在opencv3.3里，Dmatch定义成了一个类，前三个是构造函数，然后又queryIdx/trainIdx/imgIdx/distance四个值，分别表示查询图像的特征描述子索引/模板图像的特征描述子索引/有多张模板图像时的模板图像索引/匹配对间的距离。