© 2014, IJARCSSE All Rights Reserved Page | 6 Volume 4, Issue 3, March 2014 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com On the Selection of Appropriate Kernel Function for SVM in Face Recognition Justice Kwame Appati * , Gideon Kwadzo Gogovi, Gabriel Obed Fosu Department of Mathematics Kwame Nkrumah University of Science and Technology. Kumasi-Ghana, India Abstract—Biometric researchers have paid critical attention to feature extraction since without this process it is virtually impossible to perform recognition. The weaker the features, the weaker the recognition rate, likewise, the stronger the features the more likely to achieve a better recognition rate. However, the later is not always the case since after feature extraction, a classifier is required to test how robust and invariant this features extracted. This classifiers depends highly on the choice of appropriate kernel functions. We determined an appropriate kernel function suitable for the face94 database. We showed that, the selection of Multilayer Perceptron (MLP) kernel function with weight 1 and bias -1 is appropriate for this selected database since it recorded the highest recognition rate. The Support Vector Machine was used as the classifier for this research. Keywords— Kernel Function; Support Vector Machine; Biometric; Feature Extraction; Face Recognition I. INTRODUCTION Biometric, being the authentication and identification of human, based on their physiological or behavioural characteristics such as voice, face, fingerprint and gait has been in existence for quite a long time. Recent trend shows that more work is still in session as far as these characteristics are concern in the field of biometric [1 - 3]. However, each characteristic aforementioned possesses its own peculiar pros and cons and its choice is solely dependent on the problem at hand. In the past years, Vapnik and his co-workers [4] proposed the Support Vector Machines (SVMs) as an effective method for purpose pattern recognition and classification. From their proposed method, SVM finds the hyper-plane that separates the largest possible fraction of points of the same class to one side while maximizing the distance from either class to the hyper-plane, given that, the set of points belongs to two classes. According to Vapnik [4], this hyper-plane is called Optimal Separating Hyper-plane (OSH) which minimizes the risk of misclassifying not only the object in the training set but also the unseen object of the test. Osuna et al [5] train SVM for face detection, where the discrimination is between two classes: face and non-face, each with thousands of examples. Pontil and Verri [6] also use the SVMs to recognize 3D objects which are obtained from the Columbia Object Image Library (COIL) [7]. However, the appearances of these objects are explicitly different and hence the discriminations between them are not too difficult. Roobaert et al [8] repeat the experiments and argue that even a simple matching algorithm can deliver nearly the same accuracy as SVMs. They concluded that the advantage of using SVMs is not obvious. However, it is difficult to discriminate or recognize different persons (hundreds or thousands) by their faces [9] because of the similarity of faces. Guodong et al, in their research focus on the face recognition problem and show that the discrimination functions learned by SVMs can give much higher recognition accuracy than the popular standard eigenface approach [10]. They represent face images using the eigenfaces and performed feature extraction on them of which the discrimination functions between each pair are learned by SVMs. In this paper, face as biometric characteristic was considered and as can be seen, researchers have worked extensively on face feature extraction techniques and applying SVM as a classifier during recognition and classification. However, to the best of our knowledge, little has been done to investigate the effect of the various standard kernel functions available that aid the SVM to arrive at a better recognition rate during algorithm performance analysis. Since the kernel function is the heart of SVM, this paper seeks to identify an appropriate kernel function suitable for the selected face database in the application of SVM as a classifier during face recognition after feature extraction stage. II. MODEL A. Support Vector Machines Support Vector Machines (SVMs) are supervised learning algorithm for classification and is used for building models to predict whether a new object belong to a particular class or the other given a training sample with each belonging to a specified group. In a linear support vector machine, a set of N support vector z 1 , z 2 , ..., z N and weight w 1 , w 2 , ..., w N are