Face recognition using linear sparse approximation with multi-modal feature fusion

In the past few years, face recognition (FR) based on sparse representation is providing very good classification accuracy which calls for further research to exploit its capabilities to the maximum. In this paper, a novel FR method is developed as linear sparse approximation with multi-modal feature fusion (SMF classifier). SMF classifier works in three steps; multi-modal feature extraction, weight calculation for linear sparse representation and contribution calculation for final classification. In multi-modal feature extraction, a novel feature vector is developed by combining a local texture feature and a global color feature for better classification accuracy. To highlight the competency of proposed method, experiments are conducted for SMF classifier and simple sparse classifier for all possible training sets. Results are compared with simple sparse classifier in terms of mean classification accuracy. An analysis of percentage improvement in classification accuracy is performed to showcase the reliability of proposed method.