Improvement of Image Retrieval by Fusing Different Descriptors

A method for optimized image similarity measurement based on fusion of image descriptors is introduced. Descriptors of every kind can be used. The method is based on the nonparametric Wald - Wolfowitz (WW) statistical test, and the corresponding similarity measure, which is very well suited for that purpose. A simple addition of the above WW based similarity measures is the fusion rule proposed here. Experimental results in image retrieval task are very promising.