International Journal of Electrical and Computer Engineering (IJECE) Vol.2, No.6, December 2012, pp. 766~773 ISSN: 2088-8708  766 Journal homepage: http://iaesjournal.com/online/index.php/IJECE The new method of Extraction and Analysis of Non-linear Features for face recognition Ali Mahdavi Hormat*, Karim Faez**, Zeynab Shokoohi*, Mohammad Zaher Karimi* * Departement of Electrical and Computer Engineering, Qazvin Branch, Islamic Azad University ** Departement of Electrical Engineering, Amirkabir University of Technology Article Info ABSTRACT Article history: Received Jul 28, 2012 Revised Nov 19, 2012 Accepted Nov 28, 2012 In this paper, we introduce the new method of Extraction and Analysis of Non-linear Features (EANF) for face recognition based on extraction and analysis of nonlinear features i.e. Locality Preserving Analysis. In our proposed algorithm, EANF removes disadvantages such as the length of search space, different sizes and qualities of imagees due to various conditions of imaging time that has led to problems in the previous algorithms and removes the disadvantages of ELPDA methods (local neighborhood separator analysis) using the Scatter matrix in the form of a between-class scatter that this matrix introduces and displayes the nearest neighbors to K of the outer class by the samples. In addition, another advantage of EANF is high-speed in the face recognition through miniaturizing the size of feature matrix by NLPCA (Non-Linear Locality Preserving Analysis). Finally, the results of tests on FERET Dataset show the impact of the proposed method on the face recognition. Keyword: Face recognition Linear features Nonlinear features Image Processing Copyright © 2012 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: Ali mahdavi hormat, Departement of Electrical and Computer Engineering, Qazvin Branch, Islamic Azad University, Nokhbegan Blvd, Qazvin, iran. Email: ali.mahdavi.hormat@qiau.ac.ir 1. INTRODUCTION The subspace learning methods have been considered according to their position in the pattern classification of the machine vision and learning in recent years. Whithin the past two decades, many subspace learning methods have been suggested for face recognition; These methods are generally divided into two categories: The first category is the learning methods with the supervisor and the second one is without the supervisor[2-11]. In this paper, we propose a new method for the analysis of extracting nonlinear features. This method reduces the dimensions of feature matrix to 3 * N (N is the number of image samples) and also to demonstrate the problem of size in the smaller samples, our objective function includs the separator matrix of within-class scatter feature. The results performed on the FERET Dataset shows the impact of EANF method. 2. LOCAL NEIGHBORHOOD SEPARATOR ANALYSIS We consider the set of X = [x1, x2, · · ·, xN] as the example of class C {ω1, ω2, · · ·, ωC} when xi ∈ Rn. Subspace learning methods try to find the transfer function Φ, so that the transformation from n-