基于SIFT,PCA-SIFT和SURF特征检测方法的研究

Based on SIFT,PCA-SIFT and SURF Feature Detection Method Research

文章对SIFT,PCA-SIFT和SURF三种鲁棒性较强的特征检测方法作对比.文中运用KNN(K-Nearest Neighbor)和RANSAC的方法对这三种方法进行分析.其中KNN用于寻求匹配对,RANSAC用于从匹配对中剔除错误匹配.特征检测性能的鲁棒性主要是对图像旋转、图像模糊、光照变化、尺度变化下的图像进行测试.在各种图像变换中SIFT都体现出了稳定性,但计算速度相对比较慢.SURF不仅与SIFT的性能相一致,而且还拥有较快的计算速度.PCA-SIFT在图像旋转和光照变化中有较好的性能.

To study features,we compared with SIFT（Scale Invariant Feature Transform）,PCA-SIFT（Principal Component Analysis Scale Invariant Feature Transform）and SURF（Speeded Up Robust Features）three kinds of robust feature detection method.We use KNN（K-Nearest Neighbor）and random sampling method of these three kinds of methods for analysis.Where KNN used to seek matching pairs,random sampling for removing errors from match to match.Feature detection performance robustness is the image rotation,image blurring,illumination variation,the scale change of the image.The experimental evaluation is the use of repetition rate and the number of correct matching of the two statistical methods.In a variety of image transform in SIFT and SURF performance is consistent,but also has faster calculation speed.PCA-SIFT in the image rotation and illumination changes provides a better performance.