关键词: 图像分割, 活动轮廓模型, 显著性图, 拟合高斯, 水平集

Abstract: Image segmentation with intensity inhomogeneity and noise image is still a challenge in computer vision. Although existing active contour models have been demonstrated to be an effective method for image segmentation, these models are sensitive to the initial contour of evolution curve and noise, which will tend to fall into local minimum with slow speed. In order to solve these problems, this paper presents a novel algorithm based on saliency map and local likelihood fitting model. Firstly, the local likelihood fitting model is constructed by describing the neighboring intensity with local Gaussian distributions, where the means and variances of local intensity in the energy functional will vary as minimum processing. Therefore, the proposed fitting model can improve the performance of segmentation for the images with intensity inhomogeneity and noise. Secondly, the prior shape knowledge of potential segmentation object extracted by visual saliency detection method is used to initialize level set function in order to reduce the influence of initial contour selection and the time-complexity. Furthermore, the proposed model can achieve fully automatic segmentation. Experimental results demonstrate that the proposed method can provide better segmentation for the images with intensity inhomogeneity and noise when comparing the existing active contour models. The proposed method also overcomes the problems that the existing models are sensitive to the selection of initial contour and have high time-complexity.

Key words: image segmentation, active contour model, saliency map, Gaussian fitting, level set