LOCAL BINARY PATTERN BASED FEATURE EXTRACTION AND CLASSIFICATION OF FACIAL EXPRESSIONS

Abstract

Study of facial expression is a challenging problem because of facial expressions of different people are similar but not same for any particular emotion. Data mining can be applied for various tasks related to the analysis of facial expression after proper preprocessing of the images. Classification of facial expression is an important data mining technique and has many applications. Most of the existing methods of facial expression classification are based on Facial Action Coding System (FACS). The FACS is based on large number of Action Units (AUs) which are used independently of in combination form. Thus FACS encoding is very complex. Hence in the present work, Local Binary Patterns (LBP) have been used for classification of facial expressions as they make use of limited number of bins for feature extraction from images. LBP uses spatial features or micro patterns• such as spots, flat areas, corner and lines etc. We propose a new approach for facial expression classification that is based on extraction of attributes using the rotational invariant local binary pattern (RIULBP) and relative intensity in pixel groups of images. The features have been reduced using Principal Component Analysis (PCA). After reducing these features, we have used Adaptive Neuro Fuzzy Inference System (ANFIS) for classification of the facial expressions images. This approach combines the advantage of reduced and simplified process of feature extraction and ANFIS. Secondly features based on spatial aspects as well as intensity both are used together and improves the results. The experiment of this work has been done on Japanese Female Facial Expression (JAFFE) data set. Results show that the proposed RIULBP and intensity based feature extraction method and ANFIS based classification method is robust and can be used on diverse application.