Classification of Texture Images using Multiresolution Transform

Abstract: This paper proposed a scheme to identify appropriate wavelet basis function using multiresolution approach based on pyramidal wavelet transform (to extract relevant information from the texture images) for classification of the textures in various classes. In this work, for characterizing texture images at multiple scales, we have considered various wavelet basis functions such as Haar, Daubechies, Coiflet, Symlet, biorthogonal and reverse biorthogonal wavelets. They differ with each other in the formation and reconstruction. The Discrete wavelet transform is used for three level decomposition of the texture image into sub-bands. The first order statistical features are then derived from original texture image and subsequent sub-images. For classifying unknown image data to corresponding classes, minimum distance classifier is used. Experiments were carried out to compare the performance of various wavelet basis functions, which identify Haar wavelet as the best wavelet basis functions for texture classification.
.

Keywords: Texture classification, Multiresolution transform