Illumination invariant face recognition based on improved Local Binary Pattern

Local Binary Pattern (LBP) is a kind of discriminative texture descriptor for characterization of face patterns. However, the value of LBP operator is greatly changed under non-monotonic intensity transformations. As a result, the recognition performance of LBP descriptor for face images with significant illumination variations is severely dropped. In this paper, a novel illumination-invariant face recognition algorithm that applies LBP descriptor is proposed to overcome the performance degradation of LBP descriptor caused by varying illumination conditions. In our proposed algorithm, illumination variation is first compensated by the so called Dynamic Morphological Quotient Image (DMQI) which generates quotient image after morphological filtering. Then, powerful LBP operator is applied to the DMQI to derive a distinctive and robust representation for face patterns in images. We compared the recognition accuracy of the proposed algorithm with that of traditional PCA-based, LDA-based and raw LBP-based method on Yale face dataset B which contains face images with severe lighting variations. Evaluation result demonstrates that our proposed algorithm outperforms the PCA-based, LDA-based and raw LBP-based method by 22.5%, 17.4%, and 5%, respectively, in terms of the recognition accuracy on the first rank. Another advantage of our algorithm is its computational simplicity. It only takes 0.48 seconds on a Pentium IV 3.0G CPU, so it is very suitable for real-time manipulation.