Edge Detection by Anisotropic Diffusion

摘要

We present in this paper a new method for edge detection using anisotropic diffusion. We show here that edges can be robustly extracted without lost of precision and in particular at corners, even in the case of noisy images. We first present a direct method for extracting edges of the input image in "small level of noise" case. We then improve this method for "high levels of noise", proposing a new diffusion restaura-tion scheme preserving corners. We present various results on synthetic and real images.