Skip to Main Content
Proposed is an illumination normalisation approach based on the block-wise Walsh-Hadamard transform (WHT) for face recognition under illumination variations. An appropriate number of low-frequency WHT coefficients is zeroed to compensate for illumination variations. Experimental studies demonstrate that the proposed method outperforms the discrete cosine transform in the cases with larger illumination variations, while for the cases with less illumination variations the method can still obtain comparable performances with a lower computational cost. Furthermore, both analytical proof and experimental results demonstrate that the principal component analysis and the null-space-based linear discriminant analysis can be directly implemented in the WHT domain to reduce further the computational cost.