Content Based Image Retrieval Using Robust Local Octal Pattern Matrix (RLOPM)

Abstract

The local based approaches are very popular in extracting the texture features. Among these the local binary pattern (LBP) and its variants have contributed significantly in Content Based Image Retrieval (CBIR). The major disadvantage of LBP based approaches is they derive huge histogram bins and thus they derive huge dimensionality when integrated or fused with other descriptors. To address this paper put forward Robust Local Octal Pattern Matrix (RLOPM) approach. The RLOPM initially divides the 3x3 neighborhood into dual matrix of 4 sampling points each and on each sub matrix, this paper derives a robust pattern by considering the minimum of LBP and its complement pattern. This process derives a robust local octal pattern code (RLOPc) and this code ranges from 0 to 63. The derivation of co-occurrence matrix on RLOPC image derives RLOPM. The GLCM features are derived RLOPM for a precise CBIR. The experimental results on the popular databases show the efficiency of RLOPM over the existing local based approaches.