一种全局最小化的图像分割方法

A Global Minimization Method for Image Segmentation

曲线活动模型是图像分割中应用广泛且成功的一类模型,但由于能量泛函的非凸构造,使得其分割结果往往陷入局部解的困境。为了克服这一点,该文在已有的曲线活动模型之一——背景去除模型之上,从Heaviside函数的近似入手,提出了凸的能量泛函,并对其最小化,得到了相应的全局最小解求解方程。实验表明,该方法分割结果准确,分割速度快,具有一定的抗噪性,且对初始曲线的位置选取无特殊要求。

Active contour models are successfully and widely used in image segmentation. However, they always get local minima which make wrong segmentation results. In this paper, based on previous work named background removed model, a convex energy which is obtained by approximating the Heaviside function in the previous nonconvex energy is proposed. By minimizing it, the evolution equation is given. Experimental results show that the proposed method is accurate, fast and antinoise. Moreover, it is not sensitive to the location of the initial curve.