Abstract
Face recognition under unconstrained environment has various applications such as person recognition, authentication, security sensitive passport or driving license verification, face recognition on social sites. In general, pose, illumination and expression variations are the prime focused unconstrained environment for face recognition. Makeup is also one of the most challenging unconstrained environment that changes the appearance of the face due to changes in texture, shape, contrast level of mouth and eye regions. This paper proposes a robust descriptor for face recognition under makeup. We presented a color channel based local ordered maxima-minima derivative descriptor which computes the difference between consecutive maximum and minimum intensity of a local region. In the first step, color channels are separated from the makeup and non-makeup colored images. In second step, proposed descriptor ccLOMMD computes maximum and minimum difference for all pairs. Again, half generated derivative values are further used for computing maxima-minima difference for all pairs. This process is repeated until a single value achieved. This descriptor creates a discriminative feature for face recognition. Finally, Support Vector Machine is used for classification of the face images of different subjects. This descriptor performed better than state-of-the-art descriptors. In order to improve the performance, we performed the empirical study of different descriptors and fused to the proposed descriptor with a best performer Gabor based descriptor. The proposed fusion of descriptors is proved effective in the accuracy improvement, and outperformed to the existing descriptors and approaches.