GA-SVM and Mutual Information based Frequency Feature Selection for Face Recognition

The dimensionality of existing data make it difficult to deploy any information to identify features that discriminate between the classes of interest. Feature selection involves reducing the number of features, removes irrelevant, noisy and redundant data without significantly decreasing the prediction accuracy of the classifier. An efficient feature selection and classification technique for face recognition is presented in this paper. Genetic Algorithms (GAs) for feature selection and Support Vector Machine (SVM) for classification are incorporated in the proposed technique. The proposed GAs-SVM technique has two purposes in this research: Selecting of the optimal feature subset and Selecting of the kernel parameters for SVM classifier. The input feature vector for the GAs-SVM are extracted by using the Discrete Cosine Transform (DCT). We evaluate its efficiency compared to the recently proposed feature selection algorithm based on mutual information. The results show that the proposed approach is promising, it is able to select small subsets and still improve the classification accuracy.