WAVELET USE FOR IMAGE CLASSIFICATION Andrea Gavlasov´ a, Aleˇ s Proch´ azka, and Martina Mudrov´ a Prague Institute of Chemical Technology Department of Computing and Control Engineering Technick´ a 1905, 166 28 Prague 6, Czech Republic Phone: +420 - 220 444 198 ∗ Fax: +420 - 220 445 053 E-mail: Andrea.Gavlasova@vscht.cz, A.Prochazka@ieee.org, Martina.Mudrova@vscht.cz Abstract: The paper presents selected mathematical methods of image analysis including their segmentation, thresholding and feature extraction to detect specific image regions and to find their properties. The main part of the paper presents possibilities of the application of wavelet transform to find segment features using both their boundary signals and image components textures. Resulting features are then used for image segments classification by self-organizing neural networks. Proposed methods are verified for simulated image components of various sizes, rotations and textures in the Matlab environment and then used for analysis of microscopic crystal shapes and structures. Keywords: Image segmentation, distance transform, watershed transform, image features, wavelet transform, neural network classification, computational intelligence 1. INTRODUCTION Segmentation of image components, their features extraction and classification represent a specific interdisciplinary area of image processing studied in many papers and books including studies of C. et al. (2004) or M. and Aguado (2004). The main part of the paper presents the use of wavelet transform (I., 1990; E., 1994) for image features extraction using decomposition of im- age segments boundary signals or their pattern recognition. Selected features are then used for image segments classification using self-organizing neural networks (S., 1994; M., 1994). The paper presents a specific method for visualization of class boundaries based upon the work of A. and ˇ Storek (1995). All methods are verified for simulated images of different textures and sizes in the Matlab envi- ronment and then used for classification of the microscopic view of crystal shapes and their pat- terns presented in Fig. 1. Similar approach can be used for analysis of other structures including biomedical images as well. Fig. 1. Microscopic image of crystals of different shapes, textures and sizes