Doku analizinde waveletler

ABSTRACT WAVELETS FOR TEXTURE ANALYSIS KARA, Bayram M.Sc. in Electrical and Electronics Eng. Supervisor: Asst. Prof. Dr. Nurdal WATSUJI September 2003, 73 pages In this thesis, mainly, the classification of textured images which has a great importance in texture analysis has been studied. Texture analysis plays an important role in many image processing tasks such as remote sensing, medical imaging, robot vision and query by content in large image databases. An efficient classification of textured images relies on feature extraction stage. In our analysis, wavelets which can be accepted as a new analysis tool have been used. Our work is focused on derivation of appropriate wavelet coefficient properties that will lead to a successful texture analysis. It was shown that wavelet which is a multiscale representation of an image, was an appropriate tool for texture feature extraction and the statistical properties of wavelet coefficients could be used as efficient texture signatures. In this work, eight real world textured images from different natural scenes from the VisTex database were selected. Then each image was subdivided into 64 non- overlapping subimages and a database of 512 image regions of 8 texture classes was obtained. These images were analyzed with Biorthogonal Wavelets and classified with using statistical values of wavelet detail coefficients. The resulting error rates of classification experiments were compared and it was shown that, the standart, median and mean deviations of wavelet detail coefficients were appropriate features for a successful texture classification. Key words: Wavelets, texture analysis, pattern recognition, image processing. IV