A Novel Approach for Compensation of Light Variation Effects with KELM Classification for Efficient Face Recognition

Abstract

A novel technique for compensating the effect of light variations is proposed here for robust person identification using human face images. The proposed technique is adaptive and efficient for face recognition under varying illuminations generated due to light incident from different angles. Illumination variation’s effect is due to varying lighting conditions which are smoothly changing in nature. Therefore, illumination normalization is performed over some of the low-frequency discrete cosine transform (DCT) coefficients depending upon the illumination variations in the face image. A fuzzy modifier has been used to suppress the illumination variations on these low-frequency DCT coefficients. The number of low-frequency DCT coefficients is computed adaptively based upon the magnitude of these coefficients. The proposed approach utilizes KELM for the recognition of normalized face images and is tested over Extended YALE B face database. The experimental results clearly reveal that the proposed approach is significantly better than the existing approaches of illumination normalization for face recognition. With the proposed approach, the percentage error rate of 0%, 0.75%, and 1.11% on Subset 3, 4, and 5 of this database have been achieved, respectively.