Adjacent Evaluation of Completed Local Ternary Count for Texture Classification

Abstract

Local Binary Pattern (LBP) is one of the successful texture analysis methods. However, LBP suffers from noise robustness and rotation invariance. This paper proposes a novel noise insensitive texture descriptor, Adjacent Evaluation Local Ternary Count (AELTC) for rotation invariant texture classification. Unlike LBP, AELTC uses an adjacent evaluation window to change the threshold scheme. It is enhanced to Adjacent Evaluation Completed Local Ternary Count (AECLTC) with three operators to improve the performance of texture classification. During the performance evaluation, various experiments are conducted on Outex and CUReT databases using seven existing LBP variants and with proposed AECLTC. The results demonstrated the superiority of AECLTC when compared to other LBP variants.