OpenCV之SIFT/SURF算法接口介绍
早先OpenCV版本对SIFT/SURF算法可以直接调用，自OpenCV2.x之后就将SIFT/SURF等申请专利算法封装在opencv_contrib库(点集下载)函数中，需要手动下载与编译！编译过程需要CMake与Visual Studio进行，关于具体的编译过程网上存在很多教程基本都是正确的，我这里只是简单强调一下下载opencv_contrib时候需要注意事项如下图：
如上图所示，在下载opencv_contrib过程中，需要选择自己对应的版本，不然编译会出错。Ok，下满我们来看一下OpenCV中的SIFT/SURF算法对外接口函数：
SIFT/SURF算法在OpenCV源码nonfree.hpp库中，在命名空间xfeatures2d里面，下面对上图SIFT对外接口create()函数作一个简单介绍：
SURF算法对外接口如下：
SURF算法除了create()函数之外，还对外开放单独的参数设置接口：例如setHessianThreshold()、getHessianThreshold()等如上图。下面对其create()函数简单解释：
介绍完基本的SIFT/SURF算法接口信息之后，下面我们就直接上代码：
#include <iostream>
#include <opencv2\opencv.hpp>
#include <opencv2\features2d.hpp>
#include <opencv2\xfeatures2d.hpp>
#include <vector>
#include "vfc.h"
using namespace std;
using namespace cv;
int main(void)
{
// 读取输入图像
Mat img1 = imread("../opencv_SIFT/image/box.png", 0);
Mat img2 = imread("../opencv_SIFT/image/box_in_scene.png", 0);
// SIFT 特征点提取 描述子 采用默认参数
Ptr<Feature2D> feature2D = xfeatures2d::SIFT::create();
// SURF 特征点提取 描述子 采用默认参数
//Ptr<Feature2D> feature2D = xfeatures2d::SURF::create();
double features_t = (double)getTickCount();
//-- Step 1: Compute features with kpts
vector<KeyPoint> kpts1, kpts2;
feature2D->detect(img1, kpts1);
feature2D->detect(img2, kpts2);
features_t = ((double)getTickCount() - features_t) / getTickFrequency();
cout << "Features Times (s): " << features_t << " s" << endl;
double desc_t = (double)getTickCount();
//-- Step 2: Compute features' descriptors
Mat desc1, desc2;
feature2D->compute(img1, kpts1, desc1);
feature2D->compute(img2, kpts2, desc2);
// 当然，你可以直接使用detectAndCompute()函数进行特征点与描述子提取
// detectAndCompute()最后一个参数为false表示需要进行特征点提取
//feature2D->detectAndCompute(img1, Mat(), kpts1, desc1, false);
//feature2D->detectAndCompute(img2, Mat(), kpts2, desc2, false);
desc_t = ((double)getTickCount() - desc_t) / getTickFrequency();
cout << "Descriptors Times (s): " << desc_t << " s" << endl;
//////////////图像局部特征匹配BF暴力匹配或者FLANN匹配///////////////
double match_t = (double)getTickCount();
//BFMatcher matcher;
FlannBasedMatcher matcher;
vector<DMatch> matches;
matcher.match(desc1, desc2, matches);
match_t = ((double)getTickCount() - match_t) / getTickFrequency();
cout << "Match Times (s): " << match_t << " s"<< endl;
Mat showMatch;
drawMatches(img1, kpts1, img2, kpts2, matches, showMatch, Scalar::all(-1), \
Scalar::all(-1), vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);
namedWindow("RoughMatchWithOutVFC");
imshow("RoughMatchWithOutVFC", showMatch);
///////////////////////匹配点对提纯策略///////////////////////
//-- Step 4: Remove mismatches by vector field consensus (VFC)
vector<Point2f> X;
vector<Point2f> Y;
X.clear();
Y.clear();
for (unsigned int i = 0; i < matches.size(); i++) {
int idx1 = matches[i].queryIdx;
int idx2 = matches[i].trainIdx;
X.push_back(kpts1[idx1].pt);
Y.push_back(kpts2[idx2].pt);
}
// VFC process
double vfc_t = (double)getTickCount();
VFC myvfc;
myvfc.setData(X, Y);
myvfc.optimize();
vector<int> matchIdx = myvfc.obtainCorrectMatch();
vfc_t = ((double)getTickCount() - vfc_t) / getTickFrequency();
cout << "PreciseMatch Times (s): " << vfc_t << " s" << endl;
vector< DMatch > correctMatches;
vector<KeyPoint> correctKeypoints_1, correctKeypoints_2;
correctMatches.clear();
for (unsigned int i = 0; i < matchIdx.size(); i++) {
int idx = matchIdx[i];
correctMatches.push_back(matches[idx]);
correctKeypoints_1.push_back(kpts1[idx]);
correctKeypoints_2.push_back(kpts2[idx]);
}
Mat img_correctMatches;
drawMatches(img1, kpts1, img2, kpts2, correctMatches, img_correctMatches, Scalar::all(-1), \
Scalar::all(-1), vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);
namedWindow("PreciseMatchWithVFC");
imshow("PreciseMatchWithVFC", img_correctMatches);
imwrite("../opencv_SIFT/image/match.png", img_correctMatches);
waitKey(0);
return 0;
}
