水平集与深度学习融合的分割算法

Segmentation Algorithm Combining Level Set and Deep Learning

摘要: 使用水平集方法处理弱边界分割问题时，经常需要人工选择初始轮廓的位置，严重限制了水平集方法在实际中的应用。针对此问题，第一阶段使用基于小波去噪技术的级联网络做前处理，为后处理提供优秀的初始轮廓。第二阶段基于初始轮廓使用基于高斯分布的变分框架对初始轮廓进行进一步细化。实验表明此法不仅解决了手动调整初始位置的麻烦，在皮肤病数据集中表现良好。

Abstract: When using the level set method to address weak boundary segmentation problems, the initial contour position often needs to be manually selected, which severely limits the practical application of level set methods. To tackle this issue, a two-stage approach is proposed. In the first stage, a cascaded network based on wavelet denoising technology is employed for preprocessing to generate a high-quality initial contour for subsequent refinement. In the second stage, a Gaussian distribution-based variational framework is applied to further refine the initial contour. Experiments demonstrate that the proposed method not only eliminates the need for manual adjustment of the initial contour but also achieves excellent performance on dermatological image datasets.