Moment Features Weighting for Image Retrieval

Abstract

Feature selection is an effective tool to improve the performance of content based image retrieval systems. This paper presents an effective moment weighting method according to image reconstruction and retrieval accuracyto reduce the dimensionality of moment-based features. Weighting algorithms are important group of feature selection schemes. Among features employed in content based image retrieval systems, orthogonal moments are shape features which have been used to distinguish objects. But in current applications, selecting effective features among momentshas been less considered. The proposed novel weighting algorithm, obtains the weight of moment features by calculating the mean retrieval accuracy over images that are reconstructed by only one kernel coefficient and, selects the top-N features. The performance of the algorithm is compared with well-known ReliefF feature weighting and selection algorithm. Experimental results, applied on Coil-20 shape dataset, show the confidence and superiority of this feature weighing scheme over ReliefF algorithm.