An Improved eLBPH Method for Facial Identity Recognition: Expression-Specific Weighted Local Binary Pattern Histogram

Face perception is one of the most important tasks in robot vision especially for service robots. The spatially enhanced local binary pattern histogram (eLBPH) method has been proved to be effective for facial image representation and analysis, but the expression factor isn't considered and the region-dividing method is rough. In this paper, inspired by the biological mechanism of human memory and face perception, we improve the eLBPH and propose a new method, expression-specific weighted local binary pattern histogram (EWLBPH). Accordingly, the new method introduces a semantic division process and an extended modulation process into the classical eLBPH. What's more, for the facial expression recognition, we propose a novel method which utilizes the convolutional deep belief network (CDBN) to extract discriminative information and represent them effectively. Finally, through experiments we verify the rationality and effectiveness of the improvement and two psychophysical findings.