Video-Based Facial Expression Recognition Using Local Directional Binary Pattern

TL;DR

This paper introduces two advanced descriptors, VLDBP and LDBP-TOP, for improved video-based facial expression recognition, addressing lighting challenges and enhancing accuracy over traditional methods.

Contribution

It proposes novel descriptors that incorporate movement and appearance for person-independent facial expression recognition, improving accuracy over classic algorithms.

Findings

VLDBP outperforms traditional LBP in expression recognition accuracy.

LDBP-TOP simplifies computation while maintaining high performance.

Methods tested on CK+ database show significant accuracy improvements.

Abstract

Automatic facial expression analysis is a challenging issue and influenced so many areas such as human computer interaction. Due to the uncertainties of the light intensity and light direction, the face gray shades are uneven and the expression recognition rate under simple Local Binary Pattern is not ideal and promising. In this paper we propose two state-of-the-art descriptors for person-independent facial expression recognition. First the face regions of the whole images in a video sequence are modeled with Volume Local Directional Binary pattern (VLDBP), which is an extended version of the LDBP operator, incorporating movement and appearance together. To make the survey computationally simple and easy to expand, only the co-occurrences of the Local Directional Binary Pattern on three orthogonal planes (LDBP-TOP) are debated. After extracting the feature vectors the K-Nearest Neighbor classifier was used to recognize the expressions. The proposed methods are applied to the videos of the Extended Cohn-Kanade database (CK+) and the experimental outcomes demonstrate that the offered techniques achieve more accuracy in comparison with the classic and traditional algorithms.