opencv 快速线性搜索knnsearch和knnmatch比较

利用opencv的flann以及match的库进行线性近邻knn搜索的速度测试，结果表明knnmatch更快一些，看了下opencv底层代码，knnmatch用了parallel_for_ 并行加速

#include 
#include 
#include 
#include 
using namespace cv;
using namespace std;
const int samples = 50000;
const int DIM = 1024;
const int k =10;

inline double cpu_time(){
    return clock()*1000/CLOCKS_PER_SEC;
}

void generate_data(Mat& data){
    RNG rng((unsigned int)time(NULL));
    data.create(samples,DIM,CV_32FC1);

    Point center;
    center.x = rng.uniform(0,data.cols);
    center.y = rng.uniform(0,data.rows);
    rng.fill(data, RNG::NORMAL, Scalar(center.x,center.y), Scalar(data.cols*0.05,data.rows*0.05));
}

void linear_search(const Mat& data, const Mat& point, Mat& indices, Mat& dists, const int k){
    double t1,t2;
    flann::Index flannIndex(data, flann::LinearIndexParams(), cvflann::FLANN_DIST_L2);
    cout<<"Begin LinearSearch: "<64));
    t2 = cpu_time();
    cout<<"Finish linerSearch: "<"ms"<float calculate(float* a, float* b, int size, float worst_dist = -1){
    float diff0,diff1,diff2,diff3,result=0;
    float* last = a + size;
    float* lastgroup = last-3;
    while (a < lastgroup)
    {
        diff0 = a[0]-b[0];
        diff1 = a[1]-b[1];
        diff2 = a[2]-b[2];
        diff3 = a[3]-b[3];
        result += diff0*diff0 + diff1*diff1 + diff2*diff2 + diff3*diff3;

        a+=4;
        b+=4;
        if ((worst_dist>0)&&(result>worst_dist))
        {
            return result;
        }
    }
    while (areturn result;
}
//My test with loop unrolling, as it is one of the most expensive inner loops
void test(float* a, float* b){
    double t1,t2;
    t1 = cpu_time();
    cout<<"Begin test: "<float* data = a;
    float done=0;
    for(int i=0; icout<<"Finish test: "<"ms"<//Brute Force Matcher
void BFM(Mat& data, Mat& point){
    double t1, t2;
    Ptr DM = DescriptorMatcher::create("BruteForce");

    vector<vector > matches;
    cout<<"Begin BFM: "<knnMatch(point,data,matches,k);
    t2 = cpu_time();
    cout<<"Finish BFM: "<"ms"<for (int i=0;icout<0][i].trainIdx<<", ";
    }
    cout<int main(){
    Mat data;
    generate_data(data);
    Mat point = data.row(10).clone();
    float* a = (float* )data.data;
    float* b = (float* )point.data;

    test(a,b);

    Mat indices(1,DIM,CV_32FC1);
    Mat dists(1,DIM,CV_32FC1);

    BFM(data,point);

    linear_search(data,point,indices,dists,k);
    cout<return 0;
}