ASIFT算法过程实现 --- 配置避坑指南
2024-05-12 18:29:31
常规的SIFT算法进行图像匹配的时候,只能进行两个摄像机夹角比较小的(最大是15°),拍摄的图像进行相机的图像匹配,但是针对于相机之间的夹角比较大的时候,上述的算法匹配就是会出现问题.为了解决上面的这个问题,使用了一种改进的算法匹配方式ASIFT算法进行匹配.具体这种算法的优点见博客论文:基于ASIFT算法特征匹配的研究算法特征匹配的研究-图像处理文档类资源-CSDN下载在进行相机图像匹配的时候，一般的情况下我们是使用的双目相机，但是二者的视角是几乎近似平行的，平时我们更多下载资源、学习资料请访问CSDN下载频道.
https://download.csdn.net/download/m0_47489229/87108632配置过程出错:
一开始我是看的这两篇博客,进行的配置ASIFT算法在WIN7 64位系统下利用VS2012生成_zz420521的博客-CSDN博客_demo_asift欢迎大家交流，转载时请注明原文出处：http://blog.csdn.net/zlr289652981/article/details/63684857传统的sift算法在视角变化不大的情况下，可以获得良好的匹配效果，但如果在视角变化偏大时，其匹配效果会急剧下降，甚至找不到匹配点。由Guoshen Yu和Jean-Michel Morel提出的ASIFT算法很好的解决了在大视角变化条件下的图像https://blog.csdn.net/zz420521/article/details/63686163搭建可随意更改路径的VS工程-以ASIFT算法为例_zz420521的博客-CSDN博客在前一篇的博文中，笔者介绍了如何利用Guoshen Yu和Jean-Michel Morel提出的ASIFT算法的windows下C++版本源码来构建工程，具体见链接：http://blog.csdn.net/zz420521/article/details/63686163但实际上，虽然工程是跑通了，但是里面还包含了许多我们不熟悉的或者不需要的东西，尤其是一些路径设置的问题，对工程里面的设置https://blog.csdn.net/zz420521/article/details/65437441?ops_request_misc=&request_id=&biz_id=102&utm_term=vs2017%E5%AE%9E%E7%8E%B0asift%E7%AE%97%E6%B3%95&utm_medium=distribute.pc_search_result.none-task-blog-2~all~sobaiduweb~default-0-65437441.nonecase&spm=1018.2226.3001.4187我是用的VS2017进行配置,已经配置好的是OpenCV3.6.4+contribe3.6.4,然后按照上面的步骤进行配置完成之后,发现有错误出现,当然我的需求是使用代码进行编制,在进行生成解决方案的时候会出现一个如下的错误:
这里的问题貌似是需要进行提前配置Eigen和Boost库,但是我发现这个地方是直接使用一个exe文件进行操作的,并不是我所需要的.百度也没有发现想要解决的东西.
后来查阅资料发现,就是代码形式的一般是相应的python形式的,但是在查询代码的过程之中,查到一个c++形式实现这个功能的代码,如下所示:
#include <opencv2/core.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/features2d.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/calib3d.hpp>
#include <iostream>
#include <iomanip>using namespace std;
using namespace cv;static void help(char** argv)
{cout<< "This is a sample usage of AffineFeature detector/extractor.\n"<< "And this is a C++ version of samples/python/asift.py\n"<< "Usage: " << argv[0] << "\n"<< " [ --feature=<sift|orb|brisk> ] # Feature to use.\n"<< " [ --flann ] # use Flann-based matcher instead of bruteforce.\n"<< " [ --maxlines=<number(50 as default)> ] # The maximum number of lines in visualizing the matching result.\n"<< " [ --image1=<image1(aero1.jpg as default)> ]\n"<< " [ --image2=<image2(aero3.jpg as default)> ] # Path to images to compare."<< endl;
}static double timer()
{return getTickCount() / getTickFrequency();
}int main(int argc, char** argv)
{vector<String> fileName;cv::CommandLineParser parser(argc, argv,"{help h ||}""{feature|brisk|}""{flann||}""{maxlines|50|}""{image1|aero1.jpg|}{image2|aero3.jpg|}");if (parser.has("help")){help(argv);return 0;}string feature = parser.get<string>("feature");bool useFlann = parser.has("flann");int maxlines = parser.get<int>("maxlines");fileName.push_back(samples::findFile(parser.get<string>("image1")));fileName.push_back(samples::findFile(parser.get<string>("image2")));if (!parser.check()){parser.printErrors();cout << "See --help (or missing '=' between argument name and value?)" << endl;return 1;}Mat img1 = imread(fileName[0], IMREAD_GRAYSCALE);Mat img2 = imread(fileName[1], IMREAD_GRAYSCALE);if (img1.empty()){cerr << "Image " << fileName[0] << " is empty or cannot be found" << endl;return 1;}if (img2.empty()){cerr << "Image " << fileName[1] << " is empty or cannot be found" << endl;return 1;}Ptr<Feature2D> backend;Ptr<DescriptorMatcher> matcher;if (feature == "sift"){backend = SIFT::create();if (useFlann)matcher = DescriptorMatcher::create("FlannBased");elsematcher = DescriptorMatcher::create("BruteForce");}else if (feature == "orb"){backend = ORB::create();if (useFlann)matcher = makePtr<FlannBasedMatcher>(makePtr<flann::LshIndexParams>(6, 12, 1));elsematcher = DescriptorMatcher::create("BruteForce-Hamming");}else if (feature == "brisk"){backend = BRISK::create();if (useFlann)matcher = makePtr<FlannBasedMatcher>(makePtr<flann::LshIndexParams>(6, 12, 1));elsematcher = DescriptorMatcher::create("BruteForce-Hamming");}else{cerr << feature << " is not supported. See --help" << endl;return 1;}cout << "extracting with " << feature << "..." << endl;Ptr<AffineFeature> ext = AffineFeature::create(backend);vector<KeyPoint> kp1, kp2;Mat desc1, desc2;ext->detectAndCompute(img1, Mat(), kp1, desc1);ext->detectAndCompute(img2, Mat(), kp2, desc2);cout << "img1 - " << kp1.size() << " features, "<< "img2 - " << kp2.size() << " features"<< endl;cout << "matching with " << (useFlann ? "flann" : "bruteforce") << "..." << endl;double start = timer();// match and drawvector< vector<DMatch> > rawMatches;vector<Point2f> p1, p2;vector<float> distances;matcher->knnMatch(desc1, desc2, rawMatches, 2);// filter_matchesfor (size_t i = 0; i < rawMatches.size(); i++){const vector<DMatch>& m = rawMatches[i];if (m.size() == 2 && m[0].distance < m[1].distance * 0.75){p1.push_back(kp1[m[0].queryIdx].pt);p2.push_back(kp2[m[0].trainIdx].pt);distances.push_back(m[0].distance);}}vector<uchar> status;vector< pair<Point2f, Point2f> > pointPairs;Mat H = findHomography(p1, p2, status, RANSAC);int inliers = 0;for (size_t i = 0; i < status.size(); i++){if (status[i]){pointPairs.push_back(make_pair(p1[i], p2[i]));distances[inliers] = distances[i];// CV_Assert(inliers <= (int)i);inliers++;}}distances.resize(inliers);cout << "execution time: " << fixed << setprecision(2) << (timer() - start) * 1000 << " ms" << endl;cout << inliers << " / " << status.size() << " inliers/matched" << endl;cout << "visualizing..." << endl;vector<int> indices(inliers);cv::sortIdx(distances, indices, SORT_EVERY_ROW + SORT_ASCENDING);// explore_matchint h1 = img1.size().height;int w1 = img1.size().width;int h2 = img2.size().height;int w2 = img2.size().width;Mat vis = Mat::zeros(max(h1, h2), w1 + w2, CV_8U);img1.copyTo(Mat(vis, Rect(0, 0, w1, h1)));img2.copyTo(Mat(vis, Rect(w1, 0, w2, h2)));cvtColor(vis, vis, COLOR_GRAY2BGR);vector<Point2f> corners(4);corners[0] = Point2f(0, 0);corners[1] = Point2f((float)w1, 0);corners[2] = Point2f((float)w1, (float)h1);corners[3] = Point2f(0, (float)h1);vector<Point2i> icorners;perspectiveTransform(corners, corners, H);transform(corners, corners, Matx23f(1, 0, (float)w1, 0, 1, 0));Mat(corners).convertTo(icorners, CV_32S);polylines(vis, icorners, true, Scalar(255, 255, 255));for (int i = 0; i < min(inliers, maxlines); i++){int idx = indices[i];const Point2f& pi1 = pointPairs[idx].first;const Point2f& pi2 = pointPairs[idx].second;circle(vis, pi1, 2, Scalar(0, 255, 0), -1);circle(vis, pi2 + Point2f((float)w1, 0), 2, Scalar(0, 255, 0), -1);line(vis, pi1, pi2 + Point2f((float)w1, 0), Scalar(0, 255, 0));}if (inliers > maxlines)cout << "only " << maxlines << " inliers are visualized" << endl;imshow("affine find_obj", vis);waitKey();cout << "done" << endl;return 0;
}在执行的时候,发现代码报错:
还是老样子,这里的报错原因是我的库是opencv3.6.4,但是支持这个地方的是opencv4.5.0版本以上,还必须要配置contribe,所以一切还得重新弄.
我在使用配置vs2015+cmake3.24.0+opencv4.5.3+contribe4.5.3过程又出现了之前的老问题,解决方案失败,就是像下面的这个所示:
这个过程就是我是连接的外网,发现还是不行,就是需要下载很多很多的配置文件,将这些配置文件需要一个一个下载下来才可以,感觉还是比较困难的.然后就接着找博客,我发现了这一篇博客.史上最全最详细讲解 VS2015+OpenCV4.5.1+OpenCV-4.5.1_Contribute+CUDA_元宇宙MetaAI的博客-CSDN博客视频正在录制请稍等VS2015+opencv4.1+opencv4.1_contributeCmake官网CMake Warning at cmake/OpenCVDownload.cmake:202 (message)如何去掉cmake编译OpenCV时的Cmake Warning：“OpenCVGenSetupVars.cmake:54https://blog.csdn.net/CSS360/article/details/117871722?ops_request_misc=&request_id=&biz_id=102&utm_term=vs2015%20%20opencv4.5.1&utm_medium=distribute.pc_search_result.none-task-blog-2~all~sobaiduweb~default-3-117871722.142%5Ev66%5Econtrol,201%5Ev3%5Eadd_ask,213%5Ev2%5Et3_esquery_v2&spm=1018.2226.3001.4187
我的电脑已经是使用vs2017配置好别的环境了,所以只能是用vs2015进行配置新的,参考上面的操作步骤,里面的东西在他的百度网盘里面下载.cache文件添加进去,就可以了.
可以个锤子,还是不行.....................................................................................................................
前面的配置过程都是有问题的,就是我走过的坑.
第一种配置
发现要安装之前还要安装一个什么CUDA什么鬼,这具体是个什么,我现在也不知道,需要在网上进行发查找,那就接着安装...
win10 vs2015 cmake3.18.0 opencv4.5.1 opencv_contrib4.5.1 编译攻略_dragon_perfect的博客-CSDN博客_cmake编译opencv4.5.1一、前期准备：需要下载的部分有四个：OpenCV, OpenCV_contrib, CMake，VisualStudio1. 下载OpenCV and OpenCV_contrib，要求是匹配的同版本，并解压缩存储到同一文件夹下；下载OpenCV链接：https://opencv.org/releases/下载OpenCV_contrib链接 ：https://github.com/opencv/opencv_contrib/releases2. CMake的下载下载链接：ht...https://blog.csdn.net/zhulong1984/article/details/112728038https://www.baidu.com/link?url=STTSCR6UH-Zr7kAZXC8sNy3h9MSenc5VtvTHiSWMeQ30_MXa0-zrnioqai1bvwuNYjH2s4jpYwHufK5vrhn-LF0xMpJ8zL9pltVB_gINiJe&wd=&eqid=a6ef9f51000221a300000006637e2a66https://www.baidu.com/link?url=STTSCR6UH-Zr7kAZXC8sNy3h9MSenc5VtvTHiSWMeQ30_MXa0-zrnioqai1bvwuNYjH2s4jpYwHufK5vrhn-LF0xMpJ8zL9pltVB_gINiJe&wd=&eqid=a6ef9f51000221a300000006637e2a66我在看完上面两个博客的介绍之后,发现配置之中不存在什么CUDA的配置,也就是说我没有安装CUDA,那就接着CSDN,github看不懂,智力有限.那么接下来请看下面的这个博文,进行配置CUDA配置过程.
win10+cuda11.0+vs2019安装教程_离墨猫的博客-CSDN博客_cuda vs转自：https://www.jianshu.com/p/1fd15d2408bf?utm_campaign=hugo第一步：检查显卡支持的cuda版本1.第一种方法：win+R打开cmd，输入nvidia-smi，我的显卡是nvidia 2070super，支持的cuda版本是11.0图1 cmd查看显卡支持的cuda版本2.第二种方法：搜索框输入nvidia,出现nvidia控制面板，打开帮助中的系统信息，选择组件，出现cuda版本信息。第二步：安装vs2019.https://blog.csdn.net/syz201558503103/article/details/114867877?ops_request_misc=&request_id=&biz_id=102&utm_term=vs%20cuda%E5%A6%82%E4%BD%95%E5%AE%89%E8%A3%85&utm_medium=distribute.pc_search_result.none-task-blog-2~all~sobaiduweb~default-3-114867877.pc_v2_rank_dl_default&spm=1018.2226.3001.4449但是我看了一些实验室配的电脑的显卡,只有一个核显,没有独显,所以就是没有什么NVDIA配置,就没有什么CUDA配置.
到这个地方思路有断了,那么我可不可以不使用cuda的加速了.又开始了在vs2015上的配置过程,但是还是生成解决方案的时候,爆出了很多的错误.
实在没有办法了,只能在VS2017上进行配置了,突然发现可以,生成解决方案和install都没有爆出错误,很玄学. 然后我执行的代码把原来的修改了一下
#include <opencv2/core.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/features2d.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/calib3d.hpp>
#include <iostream>
#include <iomanip>using namespace std;
using namespace cv;static void help(char** argv)
{cout<< "This is a sample usage of AffineFeature detector/extractor.\n"<< "And this is a C++ version of samples/python/asift.py\n"<< "Usage: " << argv[0] << "\n"<< " [ --feature=<sift|orb|brisk> ] # Feature to use.\n"<< " [ --flann ] # use Flann-based matcher instead of bruteforce.\n"<< " [ --maxlines=<number(50 as default)> ] # The maximum number of lines in visualizing the matching result.\n"<< " [ --image1=<image1(aero1.jpg as default)> ]\n"<< " [ --image2=<image2(aero3.jpg as default)> ] # Path to images to compare."<< endl;
}static double timer()
{return getTickCount() / getTickFrequency();
}int main(int argc, char** argv)
{vector<String> fileName;cv::CommandLineParser parser(argc, argv,"{help h ||}""{feature|brisk|}""{flann||}""{maxlines|50|}""{image1|aero1.jpg|}{image2|aero3.jpg|}");if (parser.has("help")){help(argv);return 0;}string feature = parser.get<string>("feature");bool useFlann = parser.has("flann");int maxlines = parser.get<int>("maxlines");fileName.push_back(samples::findFile(parser.get<string>("image1")));fileName.push_back(samples::findFile(parser.get<string>("image2")));if (!parser.check()){parser.printErrors();cout << "See --help (or missing '=' between argument name and value?)" << endl;return 1;}Mat img1 = imread("茉莉清茶1.jpg", IMREAD_GRAYSCALE);Mat img2 = imread("茉莉清茶2.jpg", IMREAD_GRAYSCALE);if (img1.empty()){cerr << "Image1 " << " is empty or cannot be found" << endl;return 1;}if (img2.empty()){cerr << "Image2 " << " is empty or cannot be found" << endl;return 1;}Ptr<Feature2D> backend;Ptr<DescriptorMatcher> matcher;if (feature == "sift"){backend = SIFT::create();if (useFlann)matcher = DescriptorMatcher::create("FlannBased");elsematcher = DescriptorMatcher::create("BruteForce");}else if (feature == "orb"){backend = ORB::create();if (useFlann)matcher = makePtr<FlannBasedMatcher>(makePtr<flann::LshIndexParams>(6, 12, 1));elsematcher = DescriptorMatcher::create("BruteForce-Hamming");}else if (feature == "brisk"){backend = BRISK::create();if (useFlann)matcher = makePtr<FlannBasedMatcher>(makePtr<flann::LshIndexParams>(6, 12, 1));elsematcher = DescriptorMatcher::create("BruteForce-Hamming");}else{cerr << feature << " is not supported. See --help" << endl;return 1;}cout << "extracting with " << feature << "..." << endl;Ptr<AffineFeature> ext = AffineFeature::create(backend);vector<KeyPoint> kp1, kp2;Mat desc1, desc2;ext->detectAndCompute(img1, Mat(), kp1, desc1);ext->detectAndCompute(img2, Mat(), kp2, desc2);cout << "img1 - " << kp1.size() << " features, "<< "img2 - " << kp2.size() << " features"<< endl;cout << "matching with " << (useFlann ? "flann" : "bruteforce") << "..." << endl;double start = timer();// match and drawvector< vector<DMatch> > rawMatches;vector<Point2f> p1, p2;vector<float> distances;matcher->knnMatch(desc1, desc2, rawMatches, 2);// filter_matchesfor (size_t i = 0; i < rawMatches.size(); i++){const vector<DMatch>& m = rawMatches[i];if (m.size() == 2 && m[0].distance < m[1].distance * 0.75){p1.push_back(kp1[m[0].queryIdx].pt);p2.push_back(kp2[m[0].trainIdx].pt);distances.push_back(m[0].distance);}}vector<uchar> status;vector< pair<Point2f, Point2f> > pointPairs;Mat H = findHomography(p1, p2, status, RANSAC);int inliers = 0;for (size_t i = 0; i < status.size(); i++){if (status[i]){pointPairs.push_back(make_pair(p1[i], p2[i]));distances[inliers] = distances[i];// CV_Assert(inliers <= (int)i);inliers++;}}distances.resize(inliers);cout << "execution time: " << fixed << setprecision(2) << (timer() - start) * 1000 << " ms" << endl;cout << inliers << " / " << status.size() << " inliers/matched" << endl;cout << "visualizing..." << endl;vector<int> indices(inliers);cv::sortIdx(distances, indices, SORT_EVERY_ROW + SORT_ASCENDING);// explore_matchint h1 = img1.size().height;int w1 = img1.size().width;int h2 = img2.size().height;int w2 = img2.size().width;Mat vis = Mat::zeros(max(h1, h2), w1 + w2, CV_8U);img1.copyTo(Mat(vis, Rect(0, 0, w1, h1)));img2.copyTo(Mat(vis, Rect(w1, 0, w2, h2)));cvtColor(vis, vis, COLOR_GRAY2BGR);vector<Point2f> corners(4);corners[0] = Point2f(0, 0);corners[1] = Point2f((float)w1, 0);corners[2] = Point2f((float)w1, (float)h1);corners[3] = Point2f(0, (float)h1);vector<Point2i> icorners;perspectiveTransform(corners, corners, H);transform(corners, corners, Matx23f(1, 0, (float)w1, 0, 1, 0));Mat(corners).convertTo(icorners, CV_32S);polylines(vis, icorners, true, Scalar(255, 255, 255));for (int i = 0; i < min(inliers, maxlines); i++){int idx = indices[i];const Point2f& pi1 = pointPairs[idx].first;const Point2f& pi2 = pointPairs[idx].second;circle(vis, pi1, 2, Scalar(0, 255, 0), -1);circle(vis, pi2 + Point2f((float)w1, 0), 2, Scalar(0, 255, 0), -1);line(vis, pi1, pi2 + Point2f((float)w1, 0), Scalar(0, 255, 0));}if (inliers > maxlines)cout << "only " << maxlines << " inliers are visualized" << endl;imshow("affine find_obj", vis);waitKey();cout << "done" << endl;return 0;
}
貌似是没有配置相应的cuda加速,然后代码就是一直处于卡顿的状态.
因为ASIFT本来就是比较慢的,那只能慢慢等啦.反正我的CPU已经跑疯了.
这个过程是需要等待很长的时间才可以出现结果的,害,慢慢等吧.受不了了,等的时间实在太长了,等了半个小时还不行,只能是把像素降低.输出结果如下所示:
然后这个过程使用的VS2017+cmake+opencv4.5.1+contribe4.5.1(cuda没有配置),我就放在下面的百度网盘连接里了.
链接：https://pan.baidu.com/s/1hPGGiupQW-1khpiCe4QHGw?pwd=1111
提取码：1111
第二种配置
这里说的配置就是前面什么一个博客VS2012之中进行配置的一个过程,不知道为什么就是我的VS2017这个方式总是有问题,所以我使用了VS2015进行了配置.
将这个里面的代码进行改正,为什么呢?因为我看了看之前使用cmd进行操作的时候,这里会输出一定的函数,然后这里argc集合的第一个函数就是exe文件,所以第一个我就不用管,剩下的值我是自己进行配置的,由于像素的限制,就是输入的图片像素是需要进行改变的,变成一定的大小,这个是很容易实现的.后面的代码我是改成了这样的,就直接可以自己使用了.
// Copyright (c) 2008-2011, Guoshen Yu <yu@cmap.polytechnique.fr>
// Copyright (c) 2008-2011, Jean-Michel Morel <morel@cmla.ens-cachan.fr>
//
// WARNING:
// This file implements an algorithm possibly linked to the patent
//
// Jean-Michel Morel and Guoshen Yu, Method and device for the invariant
// affine recognition recognition of shapes (WO/2009/150361), patent pending.
//
// This file is made available for the exclusive aim of serving as
// scientific tool to verify of the soundness and
// completeness of the algorithm description. Compilation,
// execution and redistribution of this file may violate exclusive
// patents rights in certain countries.
// The situation being different for every country and changing
// over time, it is your responsibility to determine which patent
// rights restrictions apply to you before you compile, use,
// modify, or redistribute this file. A patent lawyer is qualified
// to make this determination.
// If and only if they don't conflict with any patent terms, you
// can benefit from the following license terms attached to this
// file.
//
// This program is provided for scientific and educational only:
// you can use and/or modify it for these purposes, but you are
// not allowed to redistribute this work or derivative works in
// source or executable form. A license must be obtained from the
// patent right holders for any other use.
//
//
//*----------------------------- demo_ASIFT --------------------------------*/
// Detect corresponding points in two images with the ASIFT method. // Please report bugs and/or send comments to Guoshen Yu yu@cmap.polytechnique.fr
//
// Reference: J.M. Morel and G.Yu, ASIFT: A New Framework for Fully Affine Invariant Image
// Comparison, SIAM Journal on Imaging Sciences, vol. 2, issue 2, pp. 438-469, 2009.
// Reference: ASIFT online demo (You can try ASIFT with your own images online.)
// http://www.ipol.im/pub/algo/my_affine_sift/
/*---------------------------------------------------------------------------*/#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <vector>
using namespace std;#ifdef _OPENMP
#include <omp.h>
#endif#include "demo_lib_sift.h"
#include "io_png/io_png.h"#include "library.h"
#include "frot.h"
#include "fproj.h"
#include "compute_asift_keypoints.h"
#include "compute_asift_matches.h"# define IM_X 800//800
# define IM_Y 600//600int main(int argc, char **argv)
{ argc = 8;cout << "?????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????" << endl;/*if ((argc != 8) && (argc != 9)) {std::cerr << " ******************************************************************************* " << std::endl<< " *************************** ASIFT image matching **************************** " << std::endl<< " ******************************************************************************* " << std::endl<< "Usage: " << argv[0] << " imgIn1.png imgIn2.png imgOutVert.png imgOutHori.png " << std::endl<< " matchings.txt keys1.txt keys2.txt [Resize option: 0/1] " << std::endl<< "- imgIn1.png, imgIn2.png: input images (in PNG format). " << std::endl<< "- imgOutVert.png, imgOutHori.png: output images (vertical/horizontal concatenated, " << std::endl<< " in PNG format.) The detected matchings are connected by write lines." << std::endl<< "- matchings.txt: coordinates of matched points (col1, row1, col2, row2). " << std::endl<< "- keys1.txt keys2.txt: ASIFT keypoints of the two images." << std::endl<< "- [optional 0/1]. 1: input images resize to 800x600 (default). 0: no resize. " << std::endl << " ******************************************************************************* " << std::endl<< " ********************* Jean-Michel Morel, Guoshen Yu, 2010 ******************** " << std::endl<< " ******************************************************************************* " << std::endl;return 1;}*/Input// Read image1argv[1] = "C:\\Users\\td\\Desktop\\xianmu\\new_stereo_match\\book3.png";argv[2] = "C:\\Users\\td\\Desktop\\xianmu\\new_stereo_match\\book4.png";argv[3] = "C:\\Users\\td\\Desktop\\xianmu\\new_stereo_match\\bookshuchu1.png";argv[4] = "C:\\Users\\td\\Desktop\\xianmu\\new_stereo_match\\bookshuchu2.png";size_t w1 , h1 ;size_t w2 , h2 ;float * iarr1;//argv[1] = "C:\\Users\\td\\Desktop\\xianmu\\new_stereo_match\\adam1.png";if (NULL == (iarr1 = read_png_f32_gray(argv[1], &w1, &h1))) {cout << iarr1 << endl;std::cerr << "Unable to load image file " << argv[1] << std::endl;return 1;}std::vector<float> ipixels1(iarr1, iarr1 + w1 * h1);free(iarr1); // Read image2float * iarr2;//argv[2] = "C:\\Users\\td\\Desktop\\xianmu\\new_stereo_match\\adam2.png";if (NULL == (iarr2 = read_png_f32_gray(argv[2], &w2, &h2))) {std::cerr << "Unable to load image file " << argv[2] << std::endl;return 1;}std::vector<float> ipixels2(iarr2, iarr2 + w2 * h2);free(iarr2); cout << "?????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????" << endl;/ Resize the images to area wS*hW in remaining the apsect-ratio / Resize if the resize flag is not set or if the flag is set unequal to 0float wS = IM_X;float hS = IM_Y;float zoom1=0, zoom2=0; int wS1=0, hS1=0, wS2=0, hS2=0;vector<float> ipixels1_zoom, ipixels2_zoom; int flag_resize = 1;if (argc == 9){ flag_resize = atoi(argv[8]);}if ((argc == 8) || (flag_resize != 0)){cout << "WARNING: The input images are resized to " << wS << "x" << hS << " for ASIFT. " << endl << " But the results will be normalized to the original image size." << endl << endl;float InitSigma_aa = 1.6;float fproj_p, fproj_bg;char fproj_i;float *fproj_x4, *fproj_y4;int fproj_o;fproj_o = 3;fproj_p = 0;fproj_i = 0;fproj_bg = 0;fproj_x4 = 0;fproj_y4 = 0;float areaS = wS * hS;// Resize image 1 float area1 = w1 * h1;zoom1 = sqrt(area1/areaS);wS1 = (int) (w1 / zoom1);hS1 = (int) (h1 / zoom1);int fproj_sx = wS1;int fproj_sy = hS1; float fproj_x1 = 0;float fproj_y1 = 0;float fproj_x2 = wS1;float fproj_y2 = 0;float fproj_x3 = 0; float fproj_y3 = hS1;/* Anti-aliasing filtering along vertical direction */if ( zoom1 > 1 ){float sigma_aa = InitSigma_aa * zoom1 / 2;GaussianBlur1D(ipixels1,w1,h1,sigma_aa,1);GaussianBlur1D(ipixels1,w1,h1,sigma_aa,0);}// simulate a tilt: subsample the image along the vertical axis by a factor of t.ipixels1_zoom.resize(wS1*hS1);fproj (ipixels1, ipixels1_zoom, w1, h1, &fproj_sx, &fproj_sy, &fproj_bg, &fproj_o, &fproj_p, &fproj_i , fproj_x1 , fproj_y1 , fproj_x2 , fproj_y2 , fproj_x3 , fproj_y3, fproj_x4, fproj_y4); // Resize image 2 float area2 = w2 * h2;zoom2 = sqrt(area2/areaS);wS2 = (int) (w2 / zoom2);hS2 = (int) (h2 / zoom2);fproj_sx = wS2;fproj_sy = hS2; fproj_x2 = wS2;fproj_y3 = hS2;/* Anti-aliasing filtering along vertical direction */if ( zoom1 > 1 ){float sigma_aa = InitSigma_aa * zoom2 / 2;GaussianBlur1D(ipixels2,w2,h2,sigma_aa,1);GaussianBlur1D(ipixels2,w2,h2,sigma_aa,0);}// simulate a tilt: subsample the image along the vertical axis by a factor of t.ipixels2_zoom.resize(wS2*hS2);fproj (ipixels2, ipixels2_zoom, w2, h2, &fproj_sx, &fproj_sy, &fproj_bg, &fproj_o, &fproj_p, &fproj_i , fproj_x1 , fproj_y1 , fproj_x2 , fproj_y2 , fproj_x3 , fproj_y3, fproj_x4, fproj_y4); }else {ipixels1_zoom.resize(w1*h1); ipixels1_zoom = ipixels1;wS1 = w1;hS1 = h1;zoom1 = 1;ipixels2_zoom.resize(w2*h2); ipixels2_zoom = ipixels2;wS2 = w2;hS2 = h2;zoom2 = 1;}/ Compute ASIFT keypoints// number N of tilts to simulate t = 1, \sqrt{2}, (\sqrt{2})^2, ..., {\sqrt{2}}^(N-1)int num_of_tilts1 = 7;int num_of_tilts2 = 7;
// int num_of_tilts1 = 1;
// int num_of_tilts2 = 1;int verb = 0;// Define the SIFT parameterssiftPar siftparameters; default_sift_parameters(siftparameters);vector< vector< keypointslist > > keys1; vector< vector< keypointslist > > keys2; int num_keys1=0, num_keys2=0;cout << "Computing keypoints on the two images..." << endl;time_t tstart, tend; tstart = time(0);num_keys1 = compute_asift_keypoints(ipixels1_zoom, wS1, hS1, num_of_tilts1, verb, keys1, siftparameters);num_keys2 = compute_asift_keypoints(ipixels2_zoom, wS2, hS2, num_of_tilts2, verb, keys2, siftparameters);tend = time(0);cout << "Keypoints computation accomplished in " << difftime(tend, tstart) << " seconds." << endl;Match ASIFT keypointsint num_matchings;matchingslist matchings; cout << "Matching the keypoints..." << endl;tstart = time(0);num_matchings = compute_asift_matches(num_of_tilts1, num_of_tilts2, wS1, hS1, wS2, hS2, verb, keys1, keys2, matchings, siftparameters);tend = time(0);cout << "Keypoints matching accomplished in " << difftime(tend, tstart) << " seconds." << endl;/ Output image containing line matches (the two images are concatenated one above the other)int band_w = 20; // insert a black band of width band_w between the two images for better visibilityint wo = MAX(w1,w2);int ho = h1+h2+band_w;float *opixelsASIFT = new float[wo*ho];for(int j = 0; j < (int) ho; j++)for(int i = 0; i < (int) wo; i++) opixelsASIFT[j*wo+i] = 255; /// Copy both images to outputfor(int j = 0; j < (int) h1; j++)for(int i = 0; i < (int) w1; i++) opixelsASIFT[j*wo+i] = ipixels1[j*w1+i]; for(int j = 0; j < (int) h2; j++)for(int i = 0; i < (int) (int)w2; i++) opixelsASIFT[(h1 + band_w + j)*wo + i] = ipixels2[j*w2 + i]; Draw matchesmatchingslist::iterator ptr = matchings.begin();for(int i=0; i < (int) matchings.size(); i++, ptr++){ draw_line(opixelsASIFT, (int) (zoom1*ptr->first.x), (int) (zoom1*ptr->first.y), (int) (zoom2*ptr->second.x), (int) (zoom2*ptr->second.y) + h1 + band_w, 255.0f, wo, ho); }/ Save imgOut write_png_f32(argv[3], opixelsASIFT, wo, ho, 1);delete[] opixelsASIFT; /*memcheck*//// Output image containing line matches (the two images are concatenated one aside the other)int woH = w1+w2+band_w;int hoH = MAX(h1,h2);float *opixelsASIFT_H = new float[woH*hoH];for(int j = 0; j < (int) hoH; j++)for(int i = 0; i < (int) woH; i++) opixelsASIFT_H[j*woH+i] = 255;/// Copy both images to outputfor(int j = 0; j < (int) h1; j++)for(int i = 0; i < (int) w1; i++) opixelsASIFT_H[j*woH+i] = ipixels1[j*w1+i]; for(int j = 0; j < (int) h2; j++)for(int i = 0; i < (int) w2; i++) opixelsASIFT_H[j*woH + w1 + band_w + i] = ipixels2[j*w2 + i]; Draw matchesmatchingslist::iterator ptrH = matchings.begin();for(int i=0; i < (int) matchings.size(); i++, ptrH++){ draw_line(opixelsASIFT_H, (int) (zoom1*ptrH->first.x), (int) (zoom1*ptrH->first.y), (int) (zoom2*ptrH->second.x) + w1 + band_w, (int) (zoom2*ptrH->second.y), 255.0f, woH, hoH); }/ Save imgOut write_png_f32(argv[4], opixelsASIFT_H, woH, hoH, 1);delete[] opixelsASIFT_H; /*memcheck*/// Write the coordinates of the matched points (row1, col1, row2, col2) to the file argv[5]std::ofstream file(argv[5]);if (file.is_open()){ // Write the number of matchings in the first linefile << num_matchings << std::endl;matchingslist::iterator ptr = matchings.begin();for(int i=0; i < (int) matchings.size(); i++, ptr++) {file << zoom1*ptr->first.x << " " << zoom1*ptr->first.y << " " << zoom2*ptr->second.x << " " << zoom2*ptr->second.y << std::endl;} }else {std::cerr << "Unable to open the file matchings."; }file.close();// Write all the keypoints (row, col, scale, orientation, desciptor (128 integers)) to // the file argv[6] (so that the users can match the keypoints with their own matching algorithm if they wish to)// keypoints in the 1st imagestd::ofstream file_key1(argv[6]);if (file_key1.is_open()){// Follow the same convention of David Lowe: // the first line contains the number of keypoints and the length of the desciptors (128)file_key1 << num_keys1 << " " << VecLength << " " << std::endl;for (int tt = 0; tt < (int) keys1.size(); tt++){for (int rr = 0; rr < (int) keys1[tt].size(); rr++){keypointslist::iterator ptr = keys1[tt][rr].begin();for(int i=0; i < (int) keys1[tt][rr].size(); i++, ptr++) {file_key1 << zoom1*ptr->x << " " << zoom1*ptr->y << " " << zoom1*ptr->scale << " " << ptr->angle;for (int ii = 0; ii < (int) VecLength; ii++){file_key1 << " " << ptr->vec[ii];}file_key1 << std::endl;}} }}else {std::cerr << "Unable to open the file keys1."; }file_key1.close();// keypoints in the 2nd imagestd::ofstream file_key2(argv[7]);if (file_key2.is_open()){// Follow the same convention of David Lowe: // the first line contains the number of keypoints and the length of the desciptors (128)file_key2 << num_keys2 << " " << VecLength << " " << std::endl;for (int tt = 0; tt < (int) keys2.size(); tt++){for (int rr = 0; rr < (int) keys2[tt].size(); rr++){keypointslist::iterator ptr = keys2[tt][rr].begin();for(int i=0; i < (int) keys2[tt][rr].size(); i++, ptr++) {file_key2 << zoom2*ptr->x << " " << zoom2*ptr->y << " " << zoom2*ptr->scale << " " << ptr->angle;for (int ii = 0; ii < (int) VecLength; ii++){file_key2 << " " << ptr->vec[ii];} file_key2 << std::endl;}} }}else {std::cerr << "Unable to open the file keys2."; }file_key2.close();return 0;
}
原始图片实在是太大了,我就把他的像素变小了一点点,然后输出的结果如下所示:
终于弄好了,太难了,花费了几天........................................
上面的两个代码我是同时运行的,总体是第一个方式比较快,需要进行降低像素.
ASIFT算法过程实现 --- 配置避坑指南相关推荐
- 算法面试避坑指南，助你轻松应对Java面试
一般来说想要成为Java开发,除了需要学习相关的技术之外,还需要学习基础知识,以便应对Java开发面试. 算法面试避坑指南五大常见算法 面试的过程其实也是一个自我认识和学习的过程,即便暂时没打算跳槽, ...
- 17条避坑指南：一份来自谷歌的数据库经验贴
点击上方蓝色"程序猿DD",选择"设为星标" 回复"资源"获取独家整理的学习资料! 来源 | https://medium.com/@rak ...
- python避坑指南_Linux下Python3.6的安装及避坑指南
Python3的安装 1.安装依赖环境 Python3在安装的过程中可能会用到各种依赖库,所以在正式安装 Python 3之前,需要将这些依赖库先行安装好. yum -y install zlib-d ...
- HarmonyOS 开发避坑指南
Harmony OS 开发避坑指南--源码下载和编译 本文介绍了如何下载鸿蒙系统源码,如何一次性配置可以编译三个目标平台(Hi3516,Hi3518和Hi3861)的编译环境,以及如何将源码编译为三个 ...
- dubbo日志关闭_不可忽视的Dubbo线程池避坑指南
推荐阅读: 阿里巴巴4面Java岗位:算法+性能调优+并发+多线程+数据库 Dubbo+Kafka+MyBatis+reids+Spring+多线程等,学完就去面试BAT 问题描述 线上突然出现Dub ...
- ext列表禁止滑动_后台列表设计避坑指南（下）
编辑导语:列表页是后台界面的重要组成之一,上篇说了后台列表设计的"搜索"设计(详情见:后台列表设计避坑指南 上):本篇继续讲剩下的两个部分的"坑"和必坑指南,我 ...
- mac下编译android源码避坑指南（新）
截至目前mac环境下android源码编译最新避坑指南 避坑方法 配置(不说配置的都是耍流氓) 下载 编译 烧录 注意事项 避坑方法 源码.SDK.机型版本一定要清楚,有些特殊的版本需要特殊的方法,官 ...
- 2022年Matlab毕设避坑指南及选题推荐
大家好,我是你的matlab大师. 2021学年,给众多的matlab方向童鞋们做了许多的课题,其中很多人加我,发现有大部分的同学因为是第一次没经验,由于自己基础不太行,在做大四的课题设计时,往往会找 ...
- Python+Selenium 网页自动化 exe 程序编程实现（最全避坑指南）
前言 在我的日常工作中,经常需要在内网(不连接互联网)的网页版办公系统中进行抓取网页数据.修改表单等大量重复性的操作.我就想是否可以编写出自动化的工具,将这些日常琐碎的操作变得轻松而高效.虽然本人非计 ...
- 【A卡，Windows】stable diffusion webui下载安装避坑指南
观前提醒 本文内容都是本人亲身经历的,一个一个安装下载测试所感,当然如果你更想用傻瓜式集成包的,那还是跳过这篇文章吧. 当然我不推荐这篇文章的操作,因为我用了差不多1h才有一副图,有N卡,就用N卡,这 ...
最新文章
- 总结和分析几种判断RecyclerView到达底部的方法
- 汇量科技收购热云数据，加速SaaS工具生态布局
- Android非常好用的组件或者框架
- 基于Spring Security的认证方式_SpringBoot认识_Spring Security OAuth2.0认证授权---springcloud工作笔记121
- Problem 1036 四塔问题
- php json 存储数据格式,文件存储（一）：通过 JSON 格式序列化文本数据
- pcap文件linux怎么打开,pcap文件用什么打开
- bin文件python读取
- MyEclipse10破解，亲测实用
- 「Python入门」Python多进程
- EXTJS记事本：当CompositeField遇上RowEditor
- 【Django下载文件-Kml文件下载】
- 国产旗舰手机定价近六千，却连iPhone12都打不过，用户选谁很明确
- Mybatis insert、update 、delete返回值
- C语言获取MTK平台系统资源信息(CPU/GPU/fps/温度等)，保存为表格形式输出
- Java——Jsoup方式的IP代理服务器的使用
- Synaptic Ubuntu软件管理工具
- 删除字符串中出现次数最少的字符，汽水瓶，简单密码
- H3CTE考试保密承诺书
- layui框架KIT ADMIN后台管理系统模板