deformable.cpp

/** @Deformable object correspondence
 ** @author Zhe Liu
 **/

/*
Copyright (C) 2011-12 Zhe Liu and Pierre Moulon.
All rights reserved.

This file is part of the KVLD library and is made available under
the terms of the BSD license (see the COPYING file).
*/

#include <algorithm>
#include <memory>

#include "kvld/kvld.h"
#include "convert.h"
#include <vector>

#include "opencv2/opencv.hpp"
#include "opencv2/features2d.hpp"

#include "opencv2/calib3d.hpp"

#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/highgui.hpp"

#include "opencv2/xfeatures2d.hpp"
#include "opencv2/xfeatures2d/nonfree.hpp"

using namespace cv::xfeatures2d;

const float sift_matching_criterion=0.98;
int main(int argc,char*argv[]) {
	std::cout<<"Warming: K-VLD may suffer performance degradation under Linux OS!"<<std::endl
	<<"Please first check existing result in the output folder as a reference!"<<std::endl;
	//================= load images ======================//
	cv::Mat image1, image2;
	int imageID=4;// index of a pair of images you want to use in folder demo_images
	std::string index;
	std::stringstream f;
	f<<imageID;
	f>>index;
	std::string input=std::string(SOURCE_DIR)+"/demo_image/IMG_";
	image1= cv::imread(input+index+".jpg", cv::IMREAD_GRAYSCALE);
	image2= cv::imread(input+index+"bis.jpg", cv::IMREAD_GRAYSCALE);

	cv::Mat image1color, image2color, concat;//for visualization
	image1color= cv::imread(input+index+".jpg", cv::IMREAD_COLOR);
	image2color= cv::imread(input+index+"bis.jpg", cv::IMREAD_COLOR);

	std::cout << image1color.rows << " " << image1color.cols << std::endl;
	std::cout << image2color.rows << " " << image2color.cols << std::endl;
	//=============== compute SIFT points =================//
	std::cout<<"Extracting SIFT features"<<std::endl;
	std::vector<cv::KeyPoint> feat1,feat2;

	cv::Ptr<SURF> detector = SURF::create(15,1);//default setting is ok, 5 levels generate too much features
	cv::Ptr<SURF> extractor= SURF::create();
	cv::Mat descriptors1,descriptors2;

	detector->detect(image1,feat1);
	extractor->compute(image1,feat1,descriptors1);
	std::cout<< "sift:: 1st image: " << feat1.size() << " keypoints"<<std::endl;

	detector->detect(image2,feat2);
	extractor->compute(image2,feat2,descriptors2);
	std::cout<< "sift:: 2nd image: " << feat2.size() << " keypoints"<<std::endl;
	//=============== compute matches using brute force matching ====================//
	std::vector<cv::DMatch> matches;
	bool bSymmetricMatches = false;//caution, activate this with knn matching will cause errors.
	cv::BFMatcher matcher(cv::NORM_L2, bSymmetricMatches);
	if (bSymmetricMatches){
		matcher.match(descriptors1,descriptors2,matches);
	}
	else
	{
		std::vector<std::vector<cv::DMatch>> knnmatches;
		matcher.knnMatch(descriptors1,descriptors2,knnmatches,2);
		for (std::vector<std::vector<cv::DMatch>>::const_iterator it=knnmatches.begin();it!=knnmatches.end();it++){
			if (it->at(0).distance<sift_matching_criterion*it->at(1).distance) 
				matches.push_back((*it)[0]);
		}
	}
	//=============== convert openCV sturctures to KVLD recognized elements
	Image<float> If1, If2;
	Convert_image(image1, If1);
	Convert_image(image2, If2);

	std::vector<keypoint> F1, F2;
	Convert_detectors(feat1,F1);//we only need detectors for KVLD
	Convert_detectors(feat2,F2);//we only need detectors for KVLD
	std::vector<Pair> matchesPair;
	Convert_matches(matches,matchesPair);

	//===============================  KVLD method ==================================//
	std::cout<<"K-VLD starts with "<<matches.size()<<" matches"<<std::endl;

	std::vector<Pair> matchesFiltered;
	std::vector<double> vec_score;

	//In order to illustrate the gvld(or vld)-consistant neighbors, the following two parameters has been externalized as inputs of the function KVLD.
	std::vector<std::vector<float>> E(matches.size(), std::vector<float>( matches.size(), -1));
	// gvld-consistency matrix, intitialized to -1, >0 consistency value, -1=unknow, -2=false  

	std::vector<bool> valide(matches.size(), true);// indices of match in the initial matches, if true at the end of KVLD, a match is kept.
	clock_t tic = clock();
	size_t it_num=0;
	KvldParameters kvldparameters;//initial parameters of KVLD
	while (it_num < 5 && kvldparameters.inlierRate>KVLD(If1, If2,F1,F2, matchesPair, matchesFiltered, vec_score,E,valide,kvldparameters)) {
		kvldparameters.inlierRate/=2;
		kvldparameters.rang_ratio=sqrt(2.0f);
		std::cout<<"low inlier rate, re-select matches with new rate="<<kvldparameters.inlierRate<<std::endl;
		if (matchesFiltered.size()==0) kvldparameters.K=2;
		it_num++;
	}
	float time_passed = ((float)clock() - tic) / CLOCKS_PER_SEC;

	std::cout<<"K-VLD filter ends with "<<matchesFiltered.size()<<" selected matches, time = "<< time_passed<<std::endl;


	//================= write files to output folder ==================//
	std::cout<<"Writing results to the output folder..."<<std::endl;
	std::string output=std::string(SOURCE_DIR)+"/demo_output/IMG_"+index+"_";
	writeResult(output,F1, F2, matchesPair, matchesFiltered, vec_score);
	//================= Visualize matching result ====================//

	cv::vconcat(image1color, image2color,concat);
	for(  std::vector<cv::DMatch>::const_iterator ptr = matches.begin(); ptr != matches.end(); ptr++)
	{
		cv::KeyPoint start = feat1[ptr->queryIdx]; 
		cv::KeyPoint end = feat2[ptr->trainIdx];
		cv::line( concat,start.pt, end.pt+cv::Point2f(0,image1.rows),cv::Scalar(0, 255,0 ));

	}
	cv::imwrite(output+"initial.png",concat);

	//========== KVLD result =============//
	cv::vconcat(image1color, image2color,concat);

	//draw gvld-consistant neighbors (not exhostive)
	for (int it1=0; it1<matchesPair.size()-1;it1++){
		for (int it2=it1+1; it2<matchesPair.size();it2++){
			if (valide[it1] && valide[it2] &&  E[it1][it2]>=0){

				cv::KeyPoint l1 = feat1[matchesPair[it1].first];
				cv::KeyPoint l2 = feat1[matchesPair[it2].first];

				cv::KeyPoint r1 = feat2[matchesPair[it1].second];
				cv::KeyPoint r2 = feat2[matchesPair[it2].second];

				cv::line(concat,l1.pt, l2.pt,cv::Scalar(255,0,255),2);
				cv::line(concat,r1.pt+cv::Point2f(0,image1.rows), r2.pt+cv::Point2f(0,image1.rows),cv::Scalar(255,0,255),2);

			}
		}
	}
	for( std::vector<Pair >::const_iterator ptr = matchesFiltered.begin(); ptr != matchesFiltered.end(); ++ptr)
	{
		size_t i = ptr->first;
		size_t j = ptr->second;
		cv::KeyPoint start = feat1[i]; 
		cv::KeyPoint end = feat2[j];
		cv::line( concat,start.pt, end.pt+cv::Point2f(0,image1.rows),cv::Scalar(0, 255, 0 ));

	}
	cv::imwrite(output+"kvld_filtered.png",concat);
	std::cout<<"Please check the output folder for results."<<std::endl;
	
	return 0;
}