Abstract: This paper analyzes the relationship between image feature types and the performance of near-duplicate image detection.Because of its good stability and discrimination，this paper chooses SIFT descriptor to detect near-duplicate images.SIFT descriptor not only is tolerant to scale changes，illumination variations，and image rotations，but also is robust to affine distortion，change of viewpoints and additive noise.Whereas，in order to overcome the shortcoming in terms of the large number of SIFT points and high computational costs for matching，this paper presents a SVD-based method to compute the similarity of two SIFT feature point sets.Experimental results show that SVD-based SIFT feature point sets matching method gets a better tradeoff between the effect of detection and detection time costs.

Key words: image local feature, Scale Invariant Feature Transform（SIFT）, near-duplicate image detection

摘要： 首先分析了不同类型的图像特征对不同重复图像类型检测性能的影响，SIFT局部描述子不仅具有良好的尺度和亮度不变性，同时对仿射形变、视角改变和噪声等也有一定的鲁棒性，因此选择了SIFT描述子来描述图像特征。同时针对SIFT特征在检测过程中匹配计算代价大的缺点，提出了基于奇异值分解的SIFT特征点集合匹配方法，实验结果表明该方法在检测效果和检测时间方面取得了一个很好的平衡。

关键词: 图像局部特征, 尺度不变特征变换（SIFT）, 重复图像检测