💥1 概述
为了提高特征点匹配的准确率，提出了一种基于改进混合滤波、特征描述符降维、SIFT特征匹配、RANSAC剔除误匹配点以及PSO算法的特征点匹配。首先将场景图像进行滤波处理达到去噪效果，然后通过特征描述符降维以减少计算量，再通过RANSAC对基于SIFT的特征点匹配进行误匹配的剔除，最后使用PSO算法进行优化以寻找到最佳的Ratio值。通过在模糊、较暗、较亮和遮挡4种以机械手为背景的场景下的图像，进行4种算法的对比实验，最后表明SIFT特征匹配算法的误匹配率最小,精确度最高。在本文章中，我们实现了哈里斯角检测器来获取与角像素相对应的兴趣点。设计用于检测图像多个比例的角落。实现了 SIFT 算法，用于获取之前找到的角点的局部特征描述符。每个角点都使用其周围图像块的梯度直方图（HoG）进行描述。使用最近距离匹配实现特征匹配，并使用 k-d 树实现 KNN 搜索。
📚2 运行结果
部分代码：
% Local Feature Stencil Code % CS 4476 / 6476: Computer Vision, Georgia Tech % Written by James Hays % This script % (1) Loads and resizes images % (2) Finds interest points in those images (you code this) % (3) Describes each interest point with a local feature (you code this) % (4) Finds matching features (you code this) % (5) Visualizes the matches % (6) Evaluates the matches based on ground truth correspondences tic close all %% 1) Load stuff % There are numerous other image sets in the supplementary data on the % project web page. You can simply download images off the Internet, as % well. However, the evaluation function at the bottom of this script will % only work for three particular image pairs (unless you add ground truth % annotations for other image pairs). It is suggested that you only work % with the two Notre Dame images until you are satisfied with your % implementation and ready to test on additional images. A single scale % pipeline works fine for these two images (and will give you full credit % for this project), but you will need local features at multiple scales to % handle harder cases. image1 = imread('../data/Notre Dame/921919841_a30df938f2_o.jpg'); image2 = imread('../data/Notre Dame/4191453057_c86028ce1f_o.jpg'); eval_file = '../data/Notre Dame/921919841_a30df938f2_o_to_4191453057_c86028ce1f_o.mat'; %image1 = imread('../data/agra_fort.jpg'); %image2 = imresize(image1,0.5); % %This pair is relatively easy (still harder than Notre Dame, though) % image1 = imread('../data/Mount Rushmore/9021235130_7c2acd9554_o.jpg'); % image2 = imread('../data/Mount Rushmore/9318872612_a255c874fb_o.jpg'); % eval_file = '../data/Mount Rushmore/9021235130_7c2acd9554_o_to_9318872612_a255c874fb_o.mat'; %This pair is relatively difficult % image1 = imread('../data/Episcopal Gaudi/4386465943_8cf9776378_o.jpg'); % image2 = imread('../data/Episcopal Gaudi/3743214471_1b5bbfda98_o.jpg'); % eval_file = '../data/Episcopal Gaudi/4386465943_8cf9776378_o_to_3743214471_1b5bbfda98_o.mat'; image1 = single(image1)/255; image2 = single(image2)/255; %make images smaller to speed up the algorithm. This parameter gets passed %into the evaluation code so don't resize the images except by changing %this parameter. scale_factor = 0.5; image1 = imresize(image1, scale_factor, 'bilinear'); image2 = imresize(image2, scale_factor, 'bilinear'); % You don't have to work with grayscale images. Matching with color % information might be helpful. image1_bw = rgb2gray(image1); image2_bw = rgb2gray(image2); feature_width = 16; %width and height of each local feature, in pixels. toc %% 2) Find distinctive points in each image. Szeliski 4.1.1 % !!! You will need to implement get_interest_points. !!! % Harris Corner Detection % [x1, y1] = get_interest_points(image1_bw, feature_width, 1); % [x2, y2] = get_interest_points(image2_bw, feature_width, 1); % % Adaptive non-maximal detection % [x1, y1] = get_interest_points_anms(image1_bw, feature_width); % toc % [x2, y2] = get_interest_points_anms(image2_bw, feature_width); % % Scaling with harris corner detection [x1, y1, confidence1, scale1] = get_interest_points_scaling(image1_bw, feature_width); [x2, y2, confidence2, scale2] = get_interest_points_scaling(image2_bw, feature_width); toc %show_correspondence(image1, image2, x1, y1, x2, y2); % % Use cheat_interest_points only for development and debugging! % [x1, y1, x2, y2] = cheat_interest_points(eval_file, scale_factor); %show_correspondence(image1, image2, x1, y1, x2, y2); %% 3) Create feature vectors at each interest point. Szeliski 4.1.2 % !!! You will need to implement get_features. !!! % SIFT % [image1_features] = get_features(image1_bw, x1, y1, feature_width); % [image2_features] = get_features(image2_bw, x2, y2, feature_width); % SIFT with scaling [image1_features] = get_features_scaling(image1_bw, x1, y1, feature_width, scale1); [image2_features] = get_features_scaling(image2_bw, x2, y2, feature_width, scale2); toc %% 4) Match features. Szeliski 4.1.3 % !!! You will need to implement get_features. !!! % % Exhaustive search [matches, confidences] = match_features(image1_features, image2_features); % % Knnsearch with kdTree %[matches, confidences] = match_features_knnsearch(image1_features, image2_features); %matches = matchFeatures(feat1, feat2); toc %% 5) Visualization % You might want to set 'num_pts_to_visualize' and 'num_pts_to_evaluate' to % some constant (e.g. 100) once you start detecting hundreds of interest % points, otherwise things might get too cluttered. You could also % threshold based on confidence. % There are two visualization functions. You can comment out one of both of % them if you prefer. num_pts_to_visualize = size(matches,1); show_correspondence(image1, image2, x1(matches(1:num_pts_to_visualize,1)), ... y1(matches(1:num_pts_to_visualize,1)), ... x2(matches(1:num_pts_to_visualize,2)), ... y2(matches(1:num_pts_to_visualize,2))); show_correspondence2(image1, image2, x1(matches(1:num_pts_to_visualize,1)), ... y1(matches(1:num_pts_to_visualize,1)), ... x2(matches(1:num_pts_to_visualize,2)), ... y2(matches(1:num_pts_to_visualize,2))); %% 6) Evaluation % This evaluation function will only work for the Notre Dame, Episcopal % Gaudi, and Mount Rushmore image pairs. Comment out this function if you % are not testing on those image pairs. Only those pairs have ground truth % available. You can use collect_ground_truth_corr.m to build the ground % truth for other image pairs if you want, but it's very tedious. It would % be a great service to the class for future years, though! num_pts_to_evaluate = size(matches,1); good_matches = evaluate_correspondence(image1, image2, eval_file, scale_factor, ... x1(matches(1:num_pts_to_evaluate,1)), ... y1(matches(1:num_pts_to_evaluate,1)), ... x2(matches(1:num_pts_to_evaluate,2)), ... y2(matches(1:num_pts_to_evaluate,2)));
🎉3 参考文献
部分理论来源于网络，如有侵权请联系删除。
[1]林曦蕾.图像局部特征匹配算法发展综述[J].现代计算机(专业版),2019(09):89-93.
[2]徐澳,华云松,夏春蕾.,陈诗雨.一种基于SIFT的改进特征点匹配算法[J].软件,2022,43(09):83-86+119.
