Face Recognition Using SVM Based on LDA Anissa Bouzalmat 1 , Jamal Kharroubi 2 and Arsalane Zarghili 3 1 Department of Computer Science faculty of Science and Technology, Sidi Mohamed Ben Abdellah University, Route d'Imouzzer Fez, 2202/30000 Morocco 2 Department of Computer Science faculty of Science and Technology, Sidi Mohamed Ben Abdellah University, Route d'Imouzzer Fez, 2202/30000 Morocco 3 Department of Computer Science faculty of Science and Technology, Sidi Mohamed Ben Abdellah University, Route d'Imouzzer Fez, 2202/30000 Morocco Abstract We present a method for face recognition that investigate the overall performance of linear ,polynomial and RBF kernel of SVM for classification based on global approach and used images having different expression variations, pose and complex backgrounds. In the first we reduce dimensional feature vector by LDA method , the result of vectors feature propagates to a set of SVM classifier, we trained SVM classifier with linear and non linear kernel for each dataset (face94, face96, grimaces)[1,2,3] in the database . Experiments demonstrate that use the LDA method combined with SVM classifier and the choice of a suitable kernel function with optimal parameters can produce high classification accuracy compared to KNN classifier on a variety of images on different Database. Keywords: Face Recognition, SVM, LDA, PCA, KNN. 1. Introduction A face recognition system recognizes a face by matching the input image against images of all faces in a database and finding the best match. It can be roughly divided into two main categories: local and global approaches. In local feature approaches a number of fiducially or control points are extracted and used for classification, while in global approaches the whole image serves as a feature vector, the techniques was developed this approach are Eigen faces ,Linear Discriminate Analysis (LDA)[4] that method outperforms PCA in terms of class discrimination, neural networks [5] and Support Vector Machines (SVM) [6], is considered easier to use and performs particularly well with high dimensional feature vectors and in case of lack of training data ,these factors which may significantly limit the performance of most neural networks [7]. The features constitute of the global image in case of the approach global are a high dimension, so it is difficult to use it without applying reduction method such high dimension of vectors have proven to be one of the biggest problems of face recognition systems then it is necessary to apply a method to reduce dimension of vectors that will perform the process for face recognition algorithm. As one of feature extraction methods for face recognition problem, linear discriminate analysis (LDA) were applied for the images this drastically reduced the number of attributes of feature vectors. By choosing the two of the most popular machine learning algorithms SVM method and K-nearest neighbors that is one of the simplest but effective in many cases in machine learning algorithms [8] then comparing the accuracy for face classification of these two classifiers, We implement a recognition system using SVM and KNN classifiers based on LDA, The procedure is described as follows: The outline of the paper is as follows: Section 2 Description of the proposed method. In section 3 contains experimental results. Section 4 concludes the paper. 2. The Proposed Method The present study proposed in this paper (Fig 1) is designed for face recognition. The system consists of: a) The whole image serves as a feature vector and reduces it by linear discriminate analysis b) Transform theses feature vectors to the format of an SVM and scale them. Finally, the obtained feature vectors are used as input of classifier support vector machine with different kernel and K-nearest neighbor classifier. IJCSI International Journal of Computer Science Issues, Vol. 10, Issue 4, No 1, July 2013 ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784 www.IJCSI.org 171 Copyright (c) 2013 International Journal of Computer Science Issues. All Rights Reserved.