Journal of Frontiers of Computer Science and Technology ›› 2013, Vol. 7 ›› Issue (12): 1115-1124.DOI: 10.3778/j.issn.1673-9418.1306007

Previous Articles     Next Articles

Chan-Vese Model with Adaptive Initial Contour for Image Segmentation

BAI Xuefei1, WANG Wenjian1,2+   

  1. 1. School of Computer and Information Technology, Shanxi University, Taiyuan 030006, China
    2. Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan 030006, China
  • Online:2013-12-01 Published:2013-12-03

自适应初始轮廓的Chan-Vese模型图像分割方法

白雪飞1,王文剑1,2+   

  1. 1. 山西大学 计算机与信息技术学院,太原 030006
    2. 山西大学 计算智能与中文信息处理教育部重点实验室,太原 030006

Abstract: Chan-Vese model (CV model) is a deformable curve moving from initial contour to object boundary under the influence of internal image force and external constraint force, which is widely used in a number of application domains including image segmentation, boundary detection and other research areas. However, little has been discussed on initial contour extracting algorithm due to the individual diversity of images. This paper proposes an adaptive CV model segmentation algorithm based on visual perception. Firstly, the priori shape information of target object regions is obtained automatically based on visual attention mechanism and image feature analyzing from bottom to top. Then, a modified CV model with self-adaptive initial contour is presented on this basis for image segmentation. The experimental results demonstrate that the proposed ACV model is robust and adaptive; therefore the impact of initial contours to active contour models is reducing effectively, meanwhile it is easier to implement and faster in computation for image segmentation than traditional CV model.

Key words: Chan-Vese model, visual perception, level set method, image segmentation

摘要: Chan-Vese模型(CV模型)是一种在图像力和外部约束力作用下从初始轮廓向目标边界运动的变形曲线,在图像分割、边缘检测等研究领域得到了广泛应用。但由于图像个体差异性较大,目前针对CV模型中初始轮廓的自动提取问题研究较少。提出了一种基于视觉认知的自适应CV模型图像分割方法。该方法根据视觉注意机制和bottom-up的底层图像特征分析,自动获取图像中目标区域的先验形状信息,用于约束CV模型中的初始轮廓,在此基础上,构造一种简化的CV模型对图像进行分割。实验结果表明,该方法具有鲁棒性和自适应性,能够有效降低初始轮廓位置对活动轮廓模型的影响,显著提高模型的收敛速度,同时减少算法迭代次数。

关键词: Chan-Vese模型, 视觉认知, 水平集方法, 图像分割