基于PDCRAN-SIFT的特征匹配技术研究

摘要

研究复杂背景下图像的目标识别,提高复杂背景下识别攻击目标的准确性和快速性.针对SIFT特征具有冗余度高、可分性差的缺点,提出PDCRAN-SIFT的方法对SIFT特征进行聚类精选,首先引入概率距离聚类的方法对SIFT特征进行聚类,选取每一类的代表特征向量作为粗聚类特征向量,然后运用随机采样一致算法剔除粗聚类特征向量中与目标图像误匹配的特征向量,从而得到精聚类匹配特征向量.实验表明,经过处理之后的PDCRAN-SIFT特征向量,冗余度大大减小,匹配时间缩短了50％左右,特征可分性和对光照、视角、噪声的鲁棒性也比SIFT特征明显增强.%This paper researches image object recognition in cluttered background to improve the accuracy and speediness of battlefield target identification.With SIFT's defects of redundancy and weak,this paper presents a PDCRAN-SIFT algorithm to improve SIFT features.First,the probabilistic distance clustering (PDC) algorithm is used to cluster the redundancy of SIFT features,select the represent features as coarse features.Then,the mismatching features are rejected by Random Sample Consensus (RANSAC) method,obtaining the refined matching features.The experiment results show the refined matching features have reduced SIFT features'redundancy to a great extent,the matching time is shortened by about 50％.What's more,they are easier to distinguish and more robust than SIFT's for light variation,rotation changes and adding noise.