基于图像引力和Mumford-Shah模型的曲线演化算法

A Algorithm of Curve Evolution Based on Image Gravitation and Mumford-Shah Model

提出了一种图像引力和Mum ford-Shah(M-S)模型的快速曲线演化算法,该算法分两步实现:首先利用图像引力,对参数主动轮廓线模型中的控制点进行寻优,使控制点能快速地收敛到图像的边缘附近;然后利用插值算法,得到目标较粗糙的大致轮廓,再使用M-S模型和窄带方法得到准确的轮廓。该算法避免了参数主动轮廓线模型不能收敛到物体凹陷的边缘和对初始位置敏感等问题,也避免了M-S模型需要对所有图像数据进行计算,计算量大等问题,实现了对参数和几何主动轮廓线模型进行了有效结合。

In the paper a method for curve evolution based on image gravitation and M-S model is presented. A twostage algorithm is thus proposed： image gravitation is first proposed to find the optima of reference points, reference points rapidly converge near image edge; The method based on narrow band level set is presented to solve the research of accurate contour. The active contour model＇s sensitivity to its initial position, the poor convergence to boundary concavities which exists in the traditional Snake and sensitivity to noise are solved. In each step of M-S iteration, the paper only deals with the data in a narrow band instead of the whole image, so the computational complexity is decreased.