Fractional Discrete Cosine Transform Based Approach for Compensating the Effect of Light Variations for Robust Face Recognition

This paper presents a novel approach of compensating the effect of light variations using fractional discrete Cosine transform (FrDCT) to efficiently solve the problem of robust person identification using face images under varying light conditions. Illumination variations are due to non-frontal lighting source with varying distribution and positions, varying ambient lighting, along with 3D shape of human face. The appearance of any object image can be characterized by illumination and reflectance generated by the object in which illumination changes slowly compared to the reflectance. Hence illumination variations mainly correspond to low frequency band of the face image. Fractional discrete Cosine transform (FrDCT) which is a generalized representation of discrete Cosine transform (DCT), has been used to process the illumination variations in the present approach. FrDcttransform the input image into $f$-domain at an angle from the input domain axis and it provides the flexibility to vary the value of the angle. In f-domain, a fuzzy filter has been employed to compensate the impact of light variations while preserving the facial features which lie in low frequency band. Percentage error rate has been used as performance metric and it is compared on Yale face database with existing state-of-art techniques of illumination normalization. The performance achieved on benchmark database clearly establishes the efficacy of the proposed approach of reducing the effect of light variations.