基于扩展级联LSH的快速影像特征匹配

In order to realize high efficient and effective image matching with the most appealing descriptor SIFT(scale invariant feature transform), An extended cascade locality sensitive hashing(LSH) matching algorithm is proposed to accelerate the image matching. First of all, a new data space floating dichotomy based hash algorithm is designed to build a high sensitive projection space, with which we achieve the accurate classification of high dimensional data thus we only need to conduct the query process within a collection of feature data with high similarity. And then, a random projection hashing is adopted to map the SIFT features into a higher dimensional Hamming space, through which the final precise calculation is conducted by combination of Hamming and Euclidean distance measurement. The experimental results indicate the matching process has been speeded up significantly while the precision is comparable to the traditional BBF and LSH based method.