Dct and Dwt Based Image Fusion for Robust Face Recognition

We propose the face recognition system using image fusion for illumination invariant robust face recognition. The proposed method is useful when less number of samples or images available in the training dataset. In our system complexity of PCA algorithm is also reduced as we are using very few (only one) fused images to train instead of more (four) images in the training dataset. Hence processing time of face recognition system also reduces as compared to the conventional method. Generally, recognition rate is better when more or sufficient training images of different illumination are available. We have used the Discrete Cosine Transform (DCT) and discrete wavelet transform (DWT) based image fusion for fusing the different illumination face images. For processing the illumination varying image, low frequency components of DCT are used to normalize the illuminated image, illumination variation, which lies mainly in the low frequency band, is normalized in the DCT domain. Other effects of illumination variation which manifest themselves by the formation of shadows and specular defects are corrected by manipulating the properties of the odd and even components of the DCT. In DWT based method the image is decomposed in high (sub-bands LH, HL and HH) and low frequencies (LL2). The approximation coefficients (low frequency), and at the same time the details are enhanced (high frequency) by averaging each element of the detail coefficient matrix. The image is reconstructed from its approximation coefficients and details coefficients in all the three directions by using the inverse wavelet transform, resulting the normalized image. With the illumination correction there is an improvement in the recognition rate, the wavelet method generates the best result, improving the recognition rate. This is due the fact that face image resultant from wavelet processes has not only enhanced contrast but also enhanced edges and detail that will facilitate the further face recognition task . _________________