自适应正则化活动轮廓模型

针对Chan-Vese模型含有许多参数,分割时需要人为调整参数,耗费大量的人力和时间的问题,提出了一个自适应正则化活动轮廓模型。首先,对Chan-Vese模型的数据项进行简化;其次,使用改进的边界加权H¹正则化代替长度项;最后,形成了一个新的不含任何参数的活动轮廓模型。在分割实验中,该模型对初始轮廓的大小、位置不敏感,具有较强的抗噪性,分割6幅图像的平均时间和迭代次数分别为1.5834 s、19次。实验结果表明,所提模型无需人工调整参数,能够分割强噪声图像和灰度不均图像,并且具有较快的分割速度。

Adaptive regularization active contour model

The Chan-Vese model for image segmentation involves many parameters, which needs to be tuned artificially for images from different modalities. The work is tedious, laborious and time-consuming. To overcome this problem, an adaptive regularization active contour model was proposed. Firstly, the data term of Chan-Vese model was reduced. Secondly, the length term was substituted by the improved edge weighted H¹ regularization term. Finally, a new active contour model was proposed without any parameters. In the segmentation experiments, the proposed model was less sensitive to the size and location of initial contour with strong noise resistance, and the average segmentation time of 6 images was 1.5834 s while the number of iterations was 19. The experimental results show that, the proposed model can handle images with intensity inhomogeneity and strong noise well without manual adjustment of parameters, and the segmentation speed is faster compared with other active contour models.