Abstract:

A K nearest neighbors detecting algorithm based on a RSOM(recursive selforganizing mapping) clustering tree was proposed by using the scale invariant feature transform(SIFT)  feature as the indices. Images were labeled and SIFT features were extracted and the number of the images were stored in the leaf node of the RSOM clustering tree. Using matched feature number as the criterion of the candidate set of K nearestneighbors set, the iterative Procrustes method was employed to obtain  more precise results. More than 50 000 images were tested and the experimental results showed the high efficiency of the proposal method.

Key words: fast algorithm, local invariant feature, content based image retrieval, recursive selforganizing mapping(RSOM), K nearest neighbonr, clustering tree, scale invariant featur tronsform(SIFT)