Local gradient increasing pattern for facial expression recognition

Abstract

This paper presents a new facial descriptor for facial expression recognition based on the Local Gradient Increasing Pattern (LGIP). A LGIP feature is to encode the intensity increasing trends in eight directions at each pixel using eight binary bits, and then a decimal code is assigned to describe the over-all increasing trend. The facial descriptor is generated from grid-based regional LGIP histograms. Subsequently, Support Vector Machine classifier is used for multi-class expression classification. Extensive experiments using Cohn-Kanade and Jaffe databases show that the LGIP based descriptor outperforms other related algorithms.