A robust algorithm for ear recognition under partial occlusion

As a promising biometrics, ear recognition is attracting increasing research interest. One challenging problem inevitable in real application is that the ears are often occluded by some objects such as hair or hat. In this paper, a general classification algorithm based on non-negative sparse representation is proposed to handle ear recognition under occlusion. Compared with conventional sparse representation in which the input data is described as a combination of basis features involving both additive and subtractive components, the proposed classification paradigm expresses an input ear signal as a linear additive combination of all the training ear signals, which is more consistent with the biological modeling of visual data. An efficient algorithm which guarantees to find the global minimum is proposed in the paper to solve the coding coefficients of the proposed system. Experimental results reveal that when the ear is partially occluded, the proposed method exhibits great robustness and achieves better recognition performance.