A novel local mesh color texture pattern for image retrieval system

Abstract

This manuscript contributes a new feature descriptor local mesh color texture pattern (LMCTP) by merging color and local spatial information for image retrieval system. The intended method combines the various features from a mixture of color spaces, the local mesh patterns of the input image. LMCTP descriptor discriminates information between the spatial color textural patterns of multi-spectral channels within the local region. It determines the relationship between surrounding neighbors of the opponent color channel. Beside this, a merged color space, I1QCb, is constructed by combining the I1 part of the I1I2I3 color space and the chromatic component images, Cb and Q, of the YCbCr and YIQ color spaces correspondingly so as to exploit the effect of joint color texture feature. The I1QCb color space, whose component of images having matching characteristics, increases the distinctive power for retrieval system. Then opponent color spaces I1Q, QCb and I1Cb are utilized for the capturing of LMCTP features. The feasibility of the proposed framework is experiments on benchmark databases such as MIT VisTex database (DB1), Corel 1000 database (DB2). Experimental study shows that the proposed pattern gives average retrieval precision of 99.8 for DB1 and 76.5 for DB2, respectively.