基于非均匀热扩散的交互式图像分割算法

交互式图像分割是图像分割中的重要分支,在现实生活和医学领域都有着广泛的应用。该文基于计算测地距离的热方法,引入了热扩散系数,提出了一种基于非均匀热扩散的交互式图像分割算法。该算法利用图像的颜色信息构造三角网格作为热扩散的媒介,首先由热方程找到距离增加的方向,再利用泊松方程还原测地距离。将前景中人工交互区域上的热流扩散速度增加,则前景不同部分之间的测地距离变小,消除了内部边界,通过设置外部边界分割限制条件,即可实现完整的前景分割。算法仅需求解两个稀疏线性方程组,鲁棒性强、精度高且更易于操作。同时,拉普拉斯算子和梯度算子的预计算可以被多次重用,减少了内存占用和时间消耗。大量交互式图像分割实验结果表明:该算法无需过多的用户交互信息,即可将现实图像中的复杂前景快速准确地分割出来。

Interactive Image Segmentation Algorithm Based on Non-uniform Heat Diffusion

Interactive image segmentation is an important branch of image segmentation and has a wide range of applications in real life and medical fields.Based on the heat method of calculating geodesic distance,we introduce the thermal diffusion coefficient and propose an interactive image segmentation algorithm based on non-uniform heat diffusion.This algorithm uses the color information of the image to construct a triangular mesh as a medium for thermal diffusion,which first finds the direction of increasing distance from the heat equation,and then uses the Poisson equation to recover the geodesic distance.Increasing the heat flow diffusion speed on the manual interaction area in the foreground reduces the geodesic distance between different parts of the foreground,eliminating the internal boundary.By setting external boundary segmentation restriction conditions,a complete foreground segmentation result can be achieved.The algorithm only needs to solve two sparse linear systems,which is robust,accurate,and simple to implement.In addition,the pre-calculation of the Laplace operator and gradient operator in the heat method can be reused multiple times,reducing memory footprint and time consumption.A large number of interactive image segmentation experiments show that the proposed algorithm can quickly and accurately segment complex foregrounds in real images without redundant user interaction information.