Texture segmentation via statistical wavelet transform modeling

In this paper, we introduce a new approach in texture segmentation utilizing 2D wavelet transform. The HL and LH subbands coefficients as the features are mapped into the probability space by 2D Generalized Gaussian probability density function (GG-PDF) to achieve preliminary segmentation. The features in PDF are classified into homogenous regions via multilevel thresholding after wavelet de-noising. The edges can be extracted from the segmented images. To verify the accuracy of GG-PDF, 2D bootstrap algorithm is used. In addition, we test our algorithm in noisy environment to check its reliability. Finally the performance of the proposed algorithm is demonstrated on variety of Bordatz textures and some textual images.