Modular FER: A Modular Facial Expression Recognition from Image Sequence Based on Two Dimensional (2D) Taylor Expansion

Abstract

This article proposes a robust facial expression recognition system, which involves two main contributions: (1) Modular approach (2) Two Dimensional Taylor Expansion for the facial feature extrication. In this facial expression recognition system (FER), the Logarithm-Laplace (LL) is used to handle illumination variation present in the face image. The modular approach mimics the human brain's capability to identify a person with limited facial regions.This article experimentally shows that some facial regions like eyes, nose, lips, and forehead contribute more to the expression recognition task. We also introduce 2D Taylor Feature Pattern (TFP) in light of the proposed Two Dimensional Taylor Expansion. Facial expression recognition's overall performance of the proposed mannequin is examined on some benchmark databases using the holistic approach and modular approach. Better recognition performance is achieved with the proposed model compared to the other existing approaches.