Modeling of the gaussian-based component analysis on the kernel space to extract face image

Abstract

One of the methods to extract the image characteristics on the feature space is Kernel Principal Component Analysis. Gaussian is a model to transform the image to the features spaces by using kernel trick. In this paper, the new model is proposed to add the image features to be more dominant, so that the main image features can be raised. Two databases were used to verify the proposed method, which are the YALE and the CAI-UTM. Three scenarios have been applied with different training samples. The results demonstrated that the proposed method can recognize the face image 87.41% for two training sets, 90.83% for three training sets, and 92.38% for four training sets on the YALE database. On the CAI-UTM database, the proposed method could classify correctly by 83.75%, 85.57%, and 87.33% for two, three, and four training sets respectively. The comparison results show that the results of the proposed approach outperformed to other methods.