A new approach for image segmentation with shape priors based on the Potts model

摘要

Shape priors play an important role for object segmentation in images with noise, distortion, shape deformation and partial occlusion. However, traditional region-based formulations often use classical level set functions, leading to complicated implementation and expensive computational costs, especially for image segmentation with multiple shape templates. To address these problems, in this paper we propose a novel segmentation formula based on the Potts model, where a reference image may contain more than one shape prior. A periodic condition and bounded region are used for the shape transformation, as we describe a new algorithm for formulation that can segment several objects simultaneously. Specifically, we focus on the use of characteristic functions as opposed to conventional classical level set functions for improved image processing efficiency. The reporting of four separate experiments using different images demonstrates the potential of the formulation and algorithm discussed.