An efficient hybrid DWT-fuzzy filter in DCT domain based illumination normalization for face recognition

This paper presents an efficient hybrid DWT-DCT based illumination normalization technique for face recognition. In a face image, illumination usually changes slowly compared to the reflectance except some casting shadows and specularities on the face. Consequently, illumination variations mainly lie in the low frequency band of the face image. Therefore, in the present work, low frequency coefficients are processed to nullify the effect of illumination variations. Discrete wavelet transform (DWT) is used to decompose the image into frequency domain. It is a sub-band coding technique which decomposes image into four sub-bands: low-low (LL), low-high (LH), high-low (HL) and high-high (HH). As illumination is related to low frequency coefficients, normalization is mainly performed on LL sub-band rather than the whole face. The fuzzy filter is applied on the appropriate number of low frequency discrete Cosine transform (DCT) coefficients of LL sub-band to minimize the variations under different lighting conditions. Also, minor corrections are performed on the rest three sub-bands. After modification, the normalized LL sub-band and rest three sub-bands are combined to generate the normalized face image. The given approach achieves zero error rates on Yale B and CMU PIE face database. Also, good performance results have been achieved on Extended Yale B face database. These results clearly confirm the effectiveness of the given approach of illumination normalization.