Circular derivative local binary pattern feature description for facial expression recognition

Abstract

This paper presents a novel feature extraction technique called circular derivative local binary pattern (CDLBP) for Facial Expression Recognition (FER). Motivated by uniform local binary patterns (uLBPs) which exhibits high discriminative potential at a reduced data dimension of the original LBP feature vector, we extract CD-LBP feature descriptors as a result of binary derivatives of the circular binary patterns formed by LBPs. Seven datasets consisting of CD-LBP feature vectors are derived from the Japanese female facial expression (JAFFE) database, fed individually in a Knearest neighbor classifier and evaluated with respect to their respective recognition rate and feature vector size. The experimental results demonstrate the relevance of the proposed feature description especially when performance metrics such as recognition accuracy and running time are considered.