Bayesian texture classification using steerable Riesz wavelets: Application to sonar images

Abstract

In this paper, the classification and segmentation of seafloor images recorded by sidescan sonar is considered. To address this problem, which can be related to texture analysis, a supervised approach based on the Bayesian framework is proposed. The features of the textured images are obtained through a parametric probabilistic model of the 2D steerable Riesz wavelet coefficients. The generalized Gaussian distribution, which is a well-established model, is used in this contribution. It is also proposed to model the approximation coefficients using the finite Gaussian mixture model to enhance the classification rate between two statistically close classes when considering only the detail coefficients. The classification results using the 2D steerable Riesz wavelets are compared to the results obtained using the classical discrete wavelets. Then, this classification method is used for image segmentation.