Modular Discriminant Feature Extraction And Its Application In Human Face Recognition

Feature extraction is a fundamental problem in the field of pattern recognition. It is the key to solve the problems such as face recognition. The work in this paper including:(1) A new method based on modular PCA and maximum scatter difference discriminate analysis (MSDDA) fusion decision is developed. In the proposed method, the original face images are divided into smaller sub-images. Then the PCA approach is applied to each of these sub-images, and the new lower dimensionality patterns take the place of the original patterns. The maximum scatter difference discriminate analysis (MSDDA) does not need to compute the inverse of the matrix, and much computational time would be saved. In the end, the maximum scatter difference discriminate analysis (MSDDA) is performed for the pattern classification. Finally, extensive experiments performed on both ORL face database and YALE face database verify the effectiveness of the proposed method.(2) Then presents a new discriminant feature extraction method which bases on modular local binary pattern. In the proposed method, the original face images are first divided into smaller sub-images. Then the LBP operator is applied to each of these sub-images and the effective texture feature is extracted. A new training set is formed by the LBP feature vector of each sub-image. The lower dimensional PCA-based features can be computed by applying PCA on the new training set. Finally, LDA is performed on the reduced PCA-based feature vectors. Experimental results on both ORL face database and YALE face database show that the proposed method is more effective than traditional PCA and LDA method which is based on the original image.(3) We develop a method called block locally principal component analysis (BLPCA) for face recognition. In the proposed method, the original face images are divided into smaller sub-images. Then the LPCA approach is applied to each of these sub-images, and the new lower dimensionality patterns take the place of the original patterns. LPCA is a linear and unsupervised subspace-learning technique, which focuses on the data points within local neighborhoods and seeks to discover the local structure of data. This local structure may contain useful information for discrimination. Combination of the main idea of modular PCA and locally principal component analysis (LPCA) may contain useful information for feature extraction. Block LPCA (BLPCA) is tested and evaluated using the ORL and YALE face database. The experimental results show that BLPCA is effective for dimension reduction and more powerful than LPCA for face recognition.