Multiscale Active Contours

摘要

In this paper, we propose an evolution equation for the active contours in scale spaces. This evolution equation is based on the Polyakov functional that has been first introduced in physics and has been then used in image processing in [17] for image denoising. Our active contours are hypersurfaces implicitly and intrinsically represented by a level set function embedded in a scale space. The scale spaces used in our approach are defined by a family of metric tensors given by the general heat diffusion equation. The well-known scale spaces such as the linear scale space, i.e. the Gaussian scale space, the Perona-Malik scale space, the mean curvature scale space and the total variation scale space can be used in this framework. A possible application of this technique is in shape analysis. For example, our multiscale segmentation technique can be coupled with the shape recognition and the shape registration algorithms to improve their robustness and their performance.