一种基于偏微分方程的图像分割算法

Partial differential equations based on the image segmentation algorithm

针对几何活动轮廓模型(GAC模型)在基于偏微分方程的图像分割领域中,算法复杂,计算量大导致演化时间长,演化速度在边界上通常不为零,引起演化曲线进入到目标的内部;或是当图像的对象有较深的凹陷边界时,曲线停在某一局部极小值状态,并不与对象的边界相一致等问题。本文提出了一种基于偏微分方程的图像分割算法,通过对停止速度场进行多尺度张量扩散,然后运用GACA模型进行分割。实验证明:本算法在不降低射线图像分割质量的前提下,可使演化时间比传统的GAC模型演化时间减少65%左右,还在一定程度上减少了边界泄露问题。

According to the geometric active contour model （ GAC model） based on partial differential equation in the field of image segmentation, the algorithm is complex, large amounts of calculation result with long evolution, evolutionary rate, on the boundary usually is not zero, leading to the curve evolution into the target internal; or when the image of the object of deep concave boundary curve, stop at a local minimum, and the boundary of the object is not consistent and other problems. This paper presents a partial differential equations based on the image segmentation algorithm, through to the halting speed field multiscale tensor diffusion, and then use the GACA model segmentation. Experiments show that： the algorithm can reduce the radiographic image segmentation under the premise of quality, can make the evolution time than the traditional GAC model evolution time reduced by about 65%, also to a certain extent reduces the boundary leaking problem.