A Rough Neurocomputing Approach for Illumination Invariant Face Recognition System

To surmount the issue of illumination variation in face recognition, this paper proposes a rough neurocomputing recognition system, namely, RNRS for an illumination invariant face recognition. The main focus of the proposed work is to address the problem of variations in illumination through the strength of the rough sets to recognize the faces under varying effects of illumination. RNRS uses geometric facial features and an approximation-decider neuron network as a recognizer. The novelty of the proposed RNRS is that the correct face match is estimated at the approximation layer itself based on the highest rough membership function value. On the contrary, if it is not being done at this layer then decider neuron does this against reduced number of sample faces. The efficiency and robustness of the proposed RNRS are demonstrated on different standard face databases and are compared with state-of-art techniques. Our proposed RNRS has achieved 93.56 % recognition rate for extended YaleB face database and 85 % recognition rate for CMU-PIE face database for larger degree of variations in illumination.