Research On Illumination Preprocessing Approaches In Face Recognition

Face recognition is one of the most successful applications in image analysis and understanding. It has received rapid development recently. However illumination variation is still one bottleneck preventing face recognition technique from being used into practice. The preprocessing approaches coping with illumination variation can be classified into two categories: the model-based approaches and the image-based approaches. Although the model-based approaches seem more perfect in theory, they commonly introduce more constraints, which make them not practical enough for applications. However, the image-based approaches commonly exploit simple and efficient image processing techniques and they are often used in practical systems. From applications, we studied the image-based approaches and achieved the following results:1. Proposed a preprocessing method based on logarithmic edges and got good results on CMU-PIE face database.2. Analyzed the current schemes validating illumination preprocessing approaches and found they have limitations and cannot validate these preprocessing approaches effectively. Based on CMU-PIE, FERET, CAS-PEAL face databases, we proposed a more general scheme to validate illumination preprocessing approaches.3. Based on the scheme proposed above, we studied and compared histogram equalization, histogram specification, Gamma intensity correction, self-quotient image, phase image, relative gradients and logarithmic edges.Based on the experimental results we got the following three conclusions:1. Some preprocessing approaches have weak generalization. Therefore, different illumination images databases should be considered in order to validate their generalization in coping with illumination variation.2. Although most preprocessing approaches can cope with illumination variation well, some may bring negative influence on images without illumination variation, so we should consider the images both with illumination variation and without illumination variation.3. When combined with different recognition algorithms, some preprocessing approaches show great difference on performance. Therefore when validating these approaches, their flexibility to different recognition algorithms should be considered.