Face Recognition Feature Extraction Methods

Face recognition is one of the popular research in the field of pattern recognition, analyzing the images containing human faces and comparing the recognized data with the human faces in the database. Face recognition is a biometric identification technology, could be widely used in all the public security system to identify human ID. In the paper, we focus on the feature analyzing methods for face recognition.Principle component analysis (PCA) approach with Karhunen-Loeve expansion could approximate lower dimensional feature vector from the original data. As the 2D image matrices must be previously transformed into ID image vectors before the PCA analysis, in this paper, we do a lot of experiments on the two-dimensional principal component analysis (2D-PCA). The size of the image covariance matrix using 2D-PCA is much smaller than PCA, and it has higher Recognition Accuracy.In order to solve the problems caused by illumination changing in PCA, we did further research on Fisher Linear Discriminant Analysis (FLD). The primary purpose of FLD is to separate samples of distinct classes by maximizing the between-class scatter while minimizing the within-class scatter. For 2D image matrix, we also use 2D-FLD to calculate the within-class scatter matrixes and between-class scatter matrixes.2D-FLD method not only avoids the colossal number of calculations, but also solves the problem about the singular of within-class scatter matrix in small sample size.Discrete Cosine Transform (DCT) could compress the information of original signal efficiently, and 2D-FLD is a simple and efficient linear projection technique for feature extraction. So we propose an approach based on DCT and 2D-FLD, the Nearest Neighbor (NN) classifier is selected to perform face recognition. Experiments demonstrate that this algorithm is better than the other methods on the computational speed and the recognition rate.