Edited by: S.Ekinović; S. Yalcin; J.Vivancos Journal of Trends in the Development of Machinery and Associated Technology Vol. 16, No. 1, 2012, ISSN 2303-4009 (online), p.p. 175-178 FACIAL EXPRESSION ANALYSIS BASED ON OPTIMIZED GABOR FEATURES Serkan Tüzün and Aydın Akan Istanbul University, Dept. of Electrical and Electronics Eng. Avcilar, 34320 Istanbul Turkey Yalçın Çekiç Bahcesehir University, Dept. of Electrical and Electronics Eng. Besiktas, 34349 Istanbul Turkey ABSTRACT In this paper, a method analyzing the facial expressions by applying Gabor filters on face images is presented. To reduce the computational complexity and dimensionality, particle swarm optimization and the mRMR methods are used. 40 Gabor filters with changing parameters are generated and the optimum one from those is determined by applying particle swarm optimization. The dimension of the Optimized Gabor Features is reduced via mRMR method based on mutual information. The facial expression recognition is performed by using only the output of optimum parameter Gabor filter as well as the all 40 filter output feature set. The performance of the proposed method is presented by means of simulations. Keywords: Facial Expression Recognition, Gabor Filters, Particle Swarm Optimization. 1. INTRODUCTION Facial expression analysis is a challenging problem, even after technical advances in signal and image processing areas. Varying illumination, pose, and partial occlusion cause large changes on the face image. The local features representing facial expressions can be extracted via spatial-frequency analysis, since it is assumed that the local features of the face are invariant to changes specified [1-3]. It is seemed that Wavelet Analysis with well localized bases in space-frequency is an appropriate choice to extract the features. Gabor functions provide the best localization both in space and in frequency, among many other wavelets [4]. Filtering face images with a Gabor wavelet is the main part of feature extraction. A Gabor wavelet consists of complex values. Because of that, Gabor features obtained by filtering contain both real and imaginary parts. Particle swarm optimization (PSO) is developed for the optimization of non-linear continuous functions, and it is inspired by bird flocking, fish schooling, and swarm theory [5]. To reduce the computational complexity and to find the optimum Gabor filter, PSO is applied on Gabor wavelet parameters. In this study we deal with a facial expression recognition method based on Gabor filter features (consisting both real and imaginary parts), and present an optimization approach for the features by using the PSO algorithm and the mRMR (minimum Redundancy Maximum Relevance) method for the dimensionality reduction.