Co-Occurrence Context of the Data-Driven Local Differential Excitation Ternary Pattern for Biomedical Image Retrieval

Abstract

This article presents a novel feature descriptor titled as local differential excitation co-occurrence pattern (LDETCoP) for bio-medical image retrieval. The proposed method aims at extracting texture features present in the medical images using differential excitation and a novel local pattern structure. All canonical pattern based methods are based on gray level difference to exploit local structure information and ignores human perception of pattern. The proposed method incorporates human perception into the pattern computation using differential excitation. Further, the proposed method exploits local structure information using radiusl (8 neighbours) and radius2 (16 neighbours) neighbouring pixels using differential excitation in an instructive way. LDETCoP also encodes co-occurrence of similar ternary edges for feature extraction. The performance of the proposed method is evaluated in terms of classification efficiency on benchmark databases VIA/I-ELCAP and NEMA-CT. The results after investigation show a substantial improvement in terms of their evaluation measures as compared to the state of the art approaches on respective databases.