Chess pattern with different weighting schemes for person independent facial expression recognition

Facial expressions are an important form of non-verbal communication as they can depict the internal mood and emotions of an individual. In Automatic Facial Expression Recognition (AFER) system, the main task is to extract features that can best classify the expressions into various categories. The existing local based approaches fail in obtaining different feature values for edge, corner and flat image regions. In this work, Chess Pattern, a game based feature descriptor is proposed based on the movements of the chessmen such as Rook, Bishop and Knight and also the combinations of Rook_Knight, Rook_Bishop and Knight_Bishop are considered for feature extraction. Apart from using binary weights, new weighting schemes such as fibonacci weights, prime weights, natural weights, squares weights are also proposed for facial feature extraction. The Chess Pattern with different weights is applied on JAFFE, MUG, TFEID, KDEF, WSEFEP and ADFES datasets for six and seven expressions. Also, for SFEW, TFEID and ADFES datasets the experiments are conducted for seven, eight and ten expressions respectively. The experiments are conducted in person independent setup, in order to simulate a real world scenario. The comparison results shows the efficiency of the proposed approach when compared to other existing methods.