曲率无关方向扩散及改进型Chan-Vese主动轮廓模型

摘要

Segmentation under low SNR is a hard problem in image segmentation research and applications. Targeting at the accuracy deficiency problem of the state-of-art non-parametric active contour models under low SNR and high curvature conditions, a new curvature-independent directional diffusion based noise-proofing method for the active contour models is proposed from the perspective of devising the noise-proofing term. The proposed method preserves active contour's shape by introducing curvature-independent directional diffusion, and furthermore distinguishes target edges from noise with unregularized edge-stopping function. Active contour's convergence to high-curvature edges is guaranteed while noise interference is suppressed effectively. Furthermore, a modified Chan-Vese active contour model is proposed for image segmentation under low SNR conditions. The effectiveness of the proposed noise-proofing method and the modified Chan-Vese active contour model are certified by thorough experiments conducted.%低信噪比条件下的图像分割是图像分割与应用中所面临的难点之一.针对当前非参数化主动轮廓分割模型在低信噪比、高曲率条件下难以准确收敛到目标边缘的问题,以主动轮廓模型中的噪声处理项作为切入点,提出一种基于曲率无关方向扩散的非参数化主动轮廓噪声去除方法.通过曲率无关方向扩散避免了去噪过程对主动轮廓形状的影响,并通过非规则边缘控制函数进一步控制噪声去除过程对信号边缘与噪声的不同作用,在有效去除噪声的同时保证了高曲率轮廓的收敛性；在此基础上,提出一种针对低信噪比图像分割的改进型Chan- Vese主动轮廓模型.最后通过详细的实验证明了该方法和改进Chan-Vese主动轮廓模型的有效性.