Variational Formulation and Multilayer Graph Based Color-Texture Image Segmentation in Multiphase

Abstract

A variational segmentation approach of color-texture images is proposed in this lecture. To improve the description ability, we use our proposed color-texture descriptor in [1] to describe color-texture. Due to heterogeneous image objects and nonlinear variation exist in color-texture image, it is not appropriate to use one single/multiple constant in Chan and Vese model for describing each phase [2-3]. Consequently, a multiphase successive active contour model (MSACM) with the multivariable Gaussian distribution is proposed to describe each phase. For geodesic active contour (GAC) has a stronger ability in capturing boundary, therefore, a newly approach incorporates the GAC with MSACM model is designed to enhance the detection ability for concave edge. As the optimization of our proposed MSACM model is NP hard problem, we cannot discrete the variational energy function of MSACM model directly, and then the multilayer graph method is adopted for getting approximate solution. To investigate the segmentation performance, lastly, a substantial of color texture images are applied for testing, and our approach achieves a significantly better performance on capture ability of concave boundary, and accuracy.