Improving the Efficiency of Face Recognition System for Images with Different Illumination Conditions
S. Ravi, M. Rethinakumari · 2011
This paper proposes a novel technique that improves the face recognition rate for multispectral face images by selecting the narrowband for the given illumination. The existing methods have demonstrated good recognition performance with frontal, centered and expressionless views of faces acquired under controlled lighting conditions. One of the main challenges to face recognition system is the degradation caused by varying illumination. In this paper, a technique that improves the recognition rate, optimal spectral range is specified automatically based on the given lighting condition of the images. This technique consists of three main sequential processes - Image acquisition, Spectral band selection and Image fusion. The given input image is compared with the gallery images, and similarity score is generated. Then quantified measure is calculated to separate the genuine and imposter similarity scores. From the previous values, optimal narrow bands are selected and images of those bands are fused together, which gives more information about the image. This fused image is fed into the classifier to calculate the recognition rate. The results produced by this method outperform all the other conventional methods used for broad-band images under varying lighting conditions.