A novel local region-based active contour model for image segmentation using Bayes theorem

Abstract

Local region-based active contour methods have been widely used to segment images with intensity inhomogeneity. However, this process can hardly segment images well when influenced by different noise. To simultaneously strengthen the anti-noise ability and preserve the distinction in segmenting images with intensity inhomogeneity, we propose characterizing image regions using local prior region descriptors under the Bayesian criterion for image segmentation. Based on the framework of Bayes theorem, a spatial regularization of connectivity maps based on a Markov random field (MRF) is introduced as the prior probability in our model. The connectivity maps enhance noise robustness by building a relationship between a pixel and its adjacent pixels. Additionally, the conditional probability of the image intensity in each local region is assumed to satisfy a Gaussian distribution with different means and deviations. Furthermore, to decrease time costs, we use the sparse field method (SFM) and compute the means and variances of the intensities in each local region before the evolution of the contour. Extensive experiments have demonstrated that the proposed method is superior to state-of-the-art active contour methods in terms of time efficiency and noise robustness.