To overcome this drawback, we research on fast algorithms for image inpainting and try to preserve edges as well as possible in this paper: Firstly，we propose an inpainting method based on level line evolution. If we implicitly represent the plane curves as the level lines of the surface, the evolution of the plane curves will lead to the evolution of the corresponding surface. The internal energy of the Snake is used as the prior model of the level line of the image, and from its descent flow we obtain the evolution equation of the image surface. The geometry meaning of this inpainting method is to connect level lines using smooth curves. Experiment shows that our method will produce better results in a comparable computing velocity, and sharp edges can be preserved well. Secondly，we propose an inpainting method using level line interpolation. It is a point by point computing method, for each pixel we can estimate it gray value by only one time computation, so it is much faster than the time match method using by the PDE method. Firstly we discuss the level line model based on the Euler’s elastic, which will be used as the local interpolation kernel to extend the level lines. Then we use the star-shaped interpolation method to interpolate the level lines layer by layer, from the boundary of the inpainting region into it. Thus we can extend the level lines smoothly into the corrupt region, and the inpainting domain will sew up step by step.
修改评论