Facial Image Analysis for Expression Recognition by Bidimensional Empirical Mode Decomposition

Abstract

Objectives: Emotion plays a significant role in non- verbal communication. The analysis of human facial images for identification of emotional states is important for producing an efficient and relevant Facial Expression Recognition System. Methods/Statistical Analysis: This paper deals with the study of different facial expression images by subtraction of facial images with its respective neutral facial image. Using Bidimensional Empirical Mode Decomposition (BEMD), the resultant subtracted images are converted into signal form by using local extrema values, viz local maxima and local minima of the image. The advantages of BEMD as compared to other techniques are, it is a full data driven method, no pre-determined filter, or no wavelet functions is required. Findings: The analysis is mainly done on eyes and mouth parts as they are the most important regions for expression recognition. The region of the eye and mouth part differs when it compared with facial image of the same person of different emotions. The difference between two images of different emotions of the same person is detected by image pixel subtraction. From the differences detected, the emotion states can be differentiated with one another and graphs are plotted for different facial expressions. Application/Improvements: The local maxima and local minima of the resultant subtracted image are plotted in the form of signal, which can be used to specify the emotional state of a facial image.