3D local ternary co-occurrence patterns for natural, texture, face and bio medical image retrieval

Abstract

In this paper, a novel feature called three-dimensional local ternary co-occurrence pattern (3D-LTCoP) is proposed for natural, texture, face and biomedical image retrieval. Standard local binary pattern and its variants like local ternary patterns, local derivative patterns, local tetra patterns etc. encode relationship between reference pixel and neighboring pixels in a two-dimensional plane of the image. The edge distribution information in these local patterns are extracted using first-order derivatives and are represented in the form of histogram. Proposed technique of feature representation draws a three-dimensional cubical image block in the local region using Gaussian filtered images and extracts relationship between reference pixel and neighboring pixels in five diverse directions of the 3D block. Further, frequency analysis of ternary patterns is performed by storing mutual local directional information in the co-occurrence matrix. Experiments are conducted on six benchmark databases ranging from natural, texture, and face to biomedical categories to observe the robustness of the proposed feature. Results are analyzed and compared with typical state-of-the-art local patterns and superiority of the proposed technique is clearly evident in terms of performance evaluation measures.