Development of feature descriptors and kernels of dimension reduction and classification for emotion recognition in human computer interaction

Abstract

Facial emotion recognition is inevitable in all human computer interactions of day-to-day life. But interpreting human emotions is a very critical task for the computers. newlineThe primary objective of this research work is to develop novel feature descriptors that aid in achieving high classification accuracy for recognizing facial emotions by a computer. To realize this objective, four feature descriptors, namely Maximum Response-based Directional Texture Pattern (MRDTP), Maximum Response-based Directional Number Pattern (MRDNP), Enhanced Gabor (E-Gabor) and Monogenic Directional Pattern (MDP) are proposed in this research work. These feature descriptors are scaling-invariant and rotation-invariant. E-Gabor obtains high classification accuracy of emotions with low dimension using an enhanced downsampling technique. Hypersphere-based normalization is proposed for normalizing the E-Gabor features. It improves the efficiency of the classification of emotions. E-Gabor features perform better after its fusion with the Pyramid Histogram of Oriented Gradients (PHOG). Both MRDTP and MRDNP are more effective than the existing directional patterns in removing random noise and providing good structural information using prominent edges which help to achieve high classification accuracy. MDP combines the directional patterns with the phase code and surface code, calculated using the first-order and second-order Riesz transforms respectively for achieving better classification accuracy. newline newline