一种基于SIFT的照片查重系统的设计与实现

Design and implementation of a photo duplicate checking system based on SIFT

传统的图像检索侧重于查找小规模图像库中的图像,对于海量图像库,其检索效率明显降低,难以提取完整的图像信息.针对上述问题,本文设计并实现了一种基于SIFT的照片查重系统.系统利用SIFT特征点四个边缘角度相对独立的特性对特征点进行分类,可大幅减少匹配过程中需要比较的特征点数量,并使用k-means算法对每一分类中的特征点进行聚类分析,然后对每一聚类的特征点进行汉明编码.匹配完成后根据特征点位置信息分析照片是否经过PS或者重组等修改.实验结果表明,在海量的图像库中进行查询时,本系统比传统的图像检索系统检索精度高,时间复杂度低.

The traditional image retrieval focuses on searching images in small-scale image library, and its retrieval efficiency obviously decreases while finding images in huge amounts of image library. It is diffi- cult to extract the complete image information. Considering above problems, this paper designs and real- izes a photo duplicate checking system based on SIFT. The system classifies the SIFT feature point u- sing their four edge Angle^s relatively independence^s feature, which can largely reduce the number of feature points need to be compared in the matching process. After that, it uses the k-means algorithm to carry cluster analysis on feature points in each category, and then hamming code is carried out for the feature point in each cluster. After the match, the article analysis whether the photos had been PS or recombined according to the feature points location. The experimental results show that the system has a high retrieval precision and low time complexity compared with the traditional image retrieval system while querying massive image library.