Reducing the effects of small sample size in DCT domain for face recognition
Saeed Dabbaghchian, Ali Aghagolzadeh, Mohammad‐Shahram Moin · 2008
The efficiency of the feature extraction approaches which are based on the statistical analysis of the training samples decreases by the small sample size. Validation of the statistical analysis depends on the number of training samples which is usually small in face recognition problems. In this paper, a statistical analysis in DCT domain is used for feature extraction, and the effect of small sample size on the feature extraction method is investigated. Finally, a simple and practical approach is proposed for decreasing the effect of small sample size. Our proposed approach divides the DCT domain (coefficients) into three bands, namely low frequency, middle frequency and high frequency bands and then utilizes the properties of each band. Simulation results on the ORL and Yale databases verify the improvement of the results by using our new approach.