Face Recognition under Varying Illumination Based on Singular Value Decomposition

Face recognition under the influence of complex illumination is a challenging problem to be solved. The common treatments for minimizing the affection of illumination variations are illumination preprocessing and illumination insensitive measure techniques. However, the methods proposed previously presents low performance. To realize high-accuracy recognition, we propose an novel illumination processing algorithm called CLAEN-SVD. Above all, singular value decomposition (SVD) is utilized to separate the face image into high-frequency and low-frequency features. Furthermore, we realize illumination normalization on the low-frequency features and enhancement on the high-frequency features via contrast limited adaptive histogram equalization (CLAHE) and threshold-value filtering, respectively. Last but not least, we reassemble the processed high-frequency and low-frequency features to form a normalized image. Experimental comparisons among our methods and some prevailing methods are put into effect on YALE B database. The experimental results demonstrate that CLAEN-SVD algorithm shows higher recognition.