Abstract: The existing local models are easy to fall into local minimum when segmenting the images with intensity inhomogeneity, which leads to the phenomenon that the evolution curve remains in the background or inside the target and can not continue to evolve and causing the segmentation failure, a model is put forward. The proposed model introduces a local intensity difference term in the energy function to drive the evolution curve across the background or inside the target by maximizing the difference between the target and the background in the neighborhood of all points on the evolution curve until it stably remains at the target edge. The experimental results show that the proposed model in this paper can effectively solve the problem caused by the local minimum in local models and improves the accuracy of complex images such as image inhomogeneity and reduces the sensitivity to the initial contour.

Key words: active contour model, local intensity difference, image segmentation, level set method, Region-Scalable Fitting（RSF） model

摘要： 针对现有局部模型在分割灰度不均匀图像时容易陷入局部极小值，导致演化曲线停留在背景处或目标内部无法继续演化从而造成分割失败的现象，提出本模型。该模型在能量泛函中增加局部灰度差异项，通过最大化演化曲线上所有点的邻域内目标和背景的差异来驱动演化曲线越过图像背景处或目标内部，直到准确地停留在目标边缘。实验结果表明提出的模型可以有效地解决局部模型因陷入局部极小值而导致的误分割问题，同时提高对分割灰度不均匀等复杂图像的准确性，并减小对初始轮廓的敏感性。

关键词: 活动轮廓模型, 局部灰度差异, 图像分割, 水平集, 区域可变的能量拟合（RSF）模型