A Two-Stage Image Segmentation Via Global and Local Region Active Contours
Hui Wang,Ting-Zhu Huang,Zhi Xu,Yugang Wang
DOI: https://doi.org/10.1016/j.neucom.2016.03.050
IF: 6
2016-01-01
Neurocomputing
Abstract:Based on popular active contours, this paper proposes a novel two-stage image segmentation method, which incorporates the global and local image region fitting energies. In the first stage, according to the global region active contour, we preliminarily segment the image by globally using the Gaussian distribution, which can rapidly get a coarse segmentation result. Subsequently, by employing a window function, we further segment the image by using the local region active contour, where we use the final active contour of the first stage as the initialization. Compared with the first stage, the local object details are accurately segmented in the second stage, which can be considered as an accurate segmentation result. Due to the suitable initialization from the first stage, the second stage works well in accurately segmenting the image, especially in local details. To regularize the level set function, we introduce a Laplace operator, which efficiently eliminates the expensive re-initialization process of traditional level set methods. Compared with the state-of-the-art methods, experiment results demonstrate the effectiveness and performance of the proposed method with applications to synthetical and real-world images, which usually contain noise, blurry boundaries, and intensity inhomogeneities.