6XSHUYLVHG3L[HO%DVHG7H[WXUH&ODVVLILFDWLRQZLWK *DERU:DYHOHW)LOWHUV Jaime Melendez 1 , Miguel Angel Garcia 2 , Domenec Puig 1 1 Intelligent Robotics and Computer Vision Group Department of Computer Science and Mathematics Rovira i Virgili University Av. Països Catalans 26, 43007 Tarragona, Spain {jaime.melendez, domenec.puig}@urv.cat 2 Department of Informatics Engineering Autonomous University of Madrid Ctra. Colmenar Viejo Km 15, 28049 Madrid, Spain {miguelangel.garcia}@uam.es $EVWUDFW This paper proposes an efficient technique for pixel-based tex- ture classification based on multichannel Gabor wavelet filters. The proposed technique is general enough to be applicable to other texture fea- ture extraction methods that also characterize the texture around image pixels through feature vectors. During the training stage, a clustering tech- nique is applied in order to compute a suitable set of prototypes that model every given texture pattern. Multisize evaluation windows are also utilized for improving the accuracy of the classifier near boundaries between regions of different texture. Experimental results with Brodatz composi- tions show the benefits of the proposed scheme in contrast with alternative approaches in terms of efficiency, memory and classification rates.  ,QWURGXFWLRQ Following the assumption that images are constituted by regions of different uniform texture patterns, texture classifiers aim at recognizing some or all of those patterns, and thereby, at identifying regions of interest in the image. In particular, pixel-based tex- ture classifiers aim at recognizing the texture patterns to which the pixels of a given image belong [1][2]. In order to accomplish this task, it is necessary to compute a set of texture features by evaluating one or more texture feature extraction methods in a neighborhood of every pixel. This neighborhood is usually defined as a square window centered at that pixel. A wide variety of texture feature extraction methods have been proposed in the lit- erature [1][3][4][5][6][7][8]. Among them, multichannel filtering techniques based on Gabor filters have received considerable attention due to some particular properties like optimal joint localization in both the spatial and frequency domains [9], and the physiological fact that 2-D Gabor filters can approximate the simple cells in the visual cortex of some mammals [10]. Typically, in a multichannel filtering scheme, an input This work has been partially supported by the Spanish Ministry of Education and Science under project DPI2004-07993-C03-03. Proceedings of the 5 th International Conference on Computer Vision Systems (ICVS 2007) Published in 2007 by Applied Computer Science Group, Bielefeld University, Germany, ISBN 978-3-00-020933-8 This document and other contributions archived and available at: http://biecoll.ub.uni-bielefeld.de