本章内容

        1. AKAZE 特征点提取
        2. 计算特征描述子
        3. FLANN特征匹配
        4. 绘制匹配结果

 

1. AKAZE 特征点提取

输出结果:

2. 单应性变换，平面对象

输出结果

源码：

#include <iostream>
#include <opencv2/opencv.hpp>
#include <opencv2/xfeatures2d.hpp>

int main(int argc, char* argv[]){
    /*本章内容
        1. AKAZE 特征点提取
        2. 计算特征描述子
        3. FLANN特征匹配
        4. 绘制匹配结果
    */

    cv::Mat src = cv::imread("/home/wang/dev/Image/book.png",cv::IMREAD_REDUCED_COLOR_2);
    cv::Mat src1 = cv::imread("/home/wang/dev/Image/books.jpg",cv::IMREAD_REDUCED_COLOR_8);

    if(src.empty() || src1.empty()){
        std::cout << "图像文件读入失败\n" <<std::endl;
        return -1;
    }
    cv::imshow("src",src);
    cv::imshow("src1",src1);
    /*1. AKAZE 特征点提取
     * api接口:  CV_WRAP static Ptr<AKAZE> create(AKAZE::DescriptorType descriptor_type = AKAZE::DESCRIPTOR_MLDB,
                                     int descriptor_size = 0, int descriptor_channels = 3,
                                     float threshold = 0.001f, int nOctaves = 4,
                                     int nOctaveLayers = 4, KAZE::DiffusivityType diffusivity = KAZE::DIFF_PM_G2);
    @param descriptor_type Type of the extracted descriptor: DESCRIPTOR_KAZE,
    DESCRIPTOR_KAZE_UPRIGHT, DESCRIPTOR_MLDB or DESCRIPTOR_MLDB_UPRIGHT.
    @param descriptor_size Size of the descriptor in bits. 0 -\> Full size
    @param descriptor_channels Number of channels in the descriptor (1, 2, 3)
    @param threshold Detector response threshold to accept point
    @param nOctaves Maximum octave evolution of the image
    @param nOctaveLayers Default number of sublevels per scale level
    @param diffusivity Diffusivity type. DIFF_PM_G1, DIFF_PM_G2, DIFF_WEICKERT or
    DIFF_CHARBONNIER
    */
    cv::Ptr<cv::AKAZE> akaze = cv::AKAZE::create();
    std::vector<cv::KeyPoint> kps1;
    akaze->detect(src,kps1);
    cv::Mat dst;
    cv::drawKeypoints(src,kps1,dst);
    cv::imshow("akaze keypoint",dst);

    // 2.计算特征描述子
    std::vector<cv::KeyPoint> kp1;
    std::vector<cv::KeyPoint> kp2;
    cv::Mat des1;
    cv::Mat des2;
    akaze->detectAndCompute(src,cv::Mat(),kp1,des1);
    akaze->detectAndCompute(src1,cv::Mat(),kp2,des2);

    // 3.FLANN特征匹配
    // LshIndexParams(int table_number, int key_size, int multi_probe_level);
    cv::flann::LshIndexParams Lsh(20, 10, 2);
    cv::FlannBasedMatcher matcher(&Lsh);
    std::vector<cv::DMatch> Dmatch;
    matcher.match(des1,des2,Dmatch);
    cv::Mat dstMatch;

    // 搜索最大最小距离
     double minDist = 1000;
     double maxDist = 0;
     for(int i=0; i < Dmatch.size();i++){
         if(Dmatch[i].distance > maxDist) maxDist = Dmatch[i].distance ;
         if(Dmatch[i].distance < minDist) minDist = Dmatch[i].distance ;
     }
     // 提取优质匹配点
     std::vector<cv::DMatch> goodMatchers;
     for(int i=0; i < Dmatch.size();i++){
         if(Dmatch[i].distance < cv::max(maxDist*0.8, 0.02)) goodMatchers.push_back(Dmatch[i]);
     }
     // 绘制匹配结果
//     cv::Mat dst;

     // 计算单应性矩阵
     std::vector<cv::Point2f> pt1s;
     std::vector<cv::Point2f> pt2s;
     for(int i=0; i < goodMatchers.size(); i++){
         pt1s.push_back(kp1[goodMatchers[i].queryIdx].pt);
         pt2s.push_back(kp2[goodMatchers[i].trainIdx].pt);
     }
     cv::Mat Hom = cv::findHomography(pt1s,pt2s,cv::RANSAC);
     std::cout << "单应性矩阵:\n" << Hom << std::endl;

     // 透视变换
     std::vector<cv::Point2f> p1;
     p1.push_back(cv::Point2f(0,0));
     p1.push_back(cv::Point2f(0,src.cols));
     p1.push_back(cv::Point2f(src.rows,src.cols));
     p1.push_back(cv::Point2f(src.rows,0));
     std::vector<cv::Point2f> p2;
     cv::perspectiveTransform(p1,p2,Hom);

     // 绘制结果
     for(int i =0; i <4 ;i++) cv::line(src1,p2[i],p2[(i+1)%4],cv::Scalar(0,255,0),4);
     cv::drawMatches(src,kp1,src1,kp2,goodMatchers,dst);
     cv::imshow("FLNN匹配结果",dst);

 

    cv::waitKey(0);
    return 1;
}