黎曼流形上内蕴方式的图像轮廓提取

Intrinsic Image Segmentation on a Riemannian Manifold

提出了一种基于黎曼几何观点的图像轮廓提取模型.在图像空间上直接赋予一种由图像灰度信息导出的黎曼度量,使之成为黎曼流形,然后在此黎曼流形上利用水平集方法对曲线以平均曲率流进行演化.由于灰度信息己嵌入黎曼流形中,演化以内蕴方式进行.计算结果表明该方法是已有模型的推广,可对曲线演化过程进行更加精细的控制.数值实验结果证实了该方法的有效性,并展示了该模型的一些特点.

An image segmentation model based on the view of Riemannian Geometry was proposed. Endowed with a Riemannian metric derived from the gray scale information of a given image, the image space became a Riemannian manifold. On this Riemannian manifold, we evolved a curve by the mean curvature flow using the level set methods. Because the gray scale information had been embedded into the Riemannian manifold, the evolution was intrinsic. Calculations show that the method is an extension of existing models; furthermore, finer controls for the evolution can be achieved. Numerical experiments show the effectivity and some good features of the proposed method.