基于扩展级联原始位置敏感散列的快速影像特征匹配  

Fast Image Feature Matching Based on Extended Cascade Original Locality Sensitive Hashing

作　　者：杨凯[1] 陈丽芳[1] 刘渊[1] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122

出　　处：《计算机工程》2016年第8期211-219,共9页Computer Engineering

基　　金：江苏省自然科学青年基金资助项目(BK20130161)

摘　　要：为解决传统基于尺度不变特征变换（SIFT）的影像匹配算法实时性较差、效率不高的问题,提出一种扩展的级联位置敏感散列（LSH）影像特征匹配算法。通过提出的数据空间浮动二分哈希构建一种比原始LSH具有更高位置敏感性的投影空间,实现对高维特征数据的划分,仅在高相似度集合中进行查询,从而提高检索速度。在各类特征集合内部进行二次随机投影散列,将特征映射到具有更好局部敏感性的高维海明空间,采用汉明距离和欧式距离相结合的测度方法,完成匹配特征对的快速查找和精确计算。实验结果表明,扩展的级联LSH影像特征匹配算法在匹配精度高于最佳Bin优先和LSH的基础上,匹配速度提高约2.5倍~3倍。In order to solve the problem of poor real-time performance and low efficiency of the traditional image matching algorithm based on Scale Invariant Feature Transform （SIFT）, an extended cascade Locality Sensitive Hashing （LSH） image feature matching algorithm is proposed. A data space floating dichotomy hash is used to build a projection space with higher locality sensitivity than the original LSH,which achieves the partition for high dimensional feature data and makes the query process only conduct in a high similarity set so as to increase the retrieval speed. Quadratic random projection hashing is adopted in all kinds of feature sets to map the features into a higher dimensional Hamming space, which has better locality sensitivity. The measurement method is used by combination of Hamming distance and Euclidean distance to complete the fast search and precise calculation. Experimental results indicate the extended cascade LSH image feature matching algorithm speeds up 2.5 times to 3 times while getting higher matching accuracy than the traditional Best Bin First（BBF） and LSH based method.

关 键 词：尺度不变特征变换 图像匹配 扩展的级联位置敏感散列 二分哈希 汉明距离 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]