Abstract:
A novel diffusion filter for low-level image processing is proposed. Analyzing the drawbacks of Gaussian convolution based regularization of partial derivatives equations, we propose an alternate method that employs anisotropic diffusion techniques to pre-smooth an image. The new technique is developed within the framework of previously proposed directional diffusion processes. Through a statistical interpretation we prove that the new filter produces consistently better results than the original version, especially when dealing with oriented textures having different spatial frequencies. Application samples are also provided in the final part of the paper.