An empirical study using combination of SVM with PSO based scattering ratio optimization and K-means

One of the most significant practical challenges for face recognition is a likeness of faces which leads to a big problem in classification of different classes. To tackle this problem, we present a novel method based on similarity of each face with other faces using the Pearson correlation coefficients. Besides, another problem is variability in lighting intensity which its physics are difficult for accurate model. In this paper, first, discrete wavelet transform (DWT) is used for feature extraction. Next, with respect to the correlation matrix, two algorithms are employed, namely, K-means clustering and particle swarm optimization (PSO) based scattering ratio matrix of correlation features. Then for each cluster, the process of classification is continued by normalization of the each subset firstly and then the decision making for each subset is performed by support vector machine (SVM). The experiments are performed on the ORL and Yale databases and the results show that there are a significant improvement in 45 features based weighted recognition rate.