Abstract: C-V model is an effective segmentation method for medical brain tumor images. Aimed at its sensitivity to the initial contour position and time-consuming re-initialization, a new method for C-V model optimized by watershed transformation is proposed. First, the marker function is introduced, with mandatory minimum technology, to deal with the over-segmentation of traditional watershed transformation and get the coarse result. Then, the initial contour position of C-V model is confirmed based on the coarse segmentation. Finally, the C-V model without re-initialization is used for fine segmentation, obtaining the accurate segmentation. The experimental results show that the proposed method can effectively segment the common brain tumor, but also for the tumor which is attached to tissues.

Key words:  C-V model, image segmentation, watershed transformation, brain tumor image

摘要： C-V模型可有效对脑肿瘤等医学图像进行分割，但存在对初始轮廓位置敏感及重新初始化耗时的问题，为此，提出了一种分水岭优化的C-V模型脑肿瘤分割方法。首先引入标记函数，通过强制最小技术改善传统分水岭变换的过分割现象，得到粗分割结果，然后在粗分割基础上确定C-V模型初始轮廓位置，最后采用无需重新初始化的C-V模型进行细分割，得到较精确的脑肿瘤分割结果。实例结果表明，经过分水岭优化后的C-V模型能够对常见脑肿瘤图像进行有效分割，尤其是能够将与组织粘连的肿瘤分割出来。

关键词: C-V模型, 图像分割, 分水岭变换, 脑肿瘤图像