Blocking artifacts reduction for DCT-based image compression using neurofuzzy driven anisotropic diffusion

摘要

In this paper, we propose a new deblocking method for DCT-based compressed images using anisotropic diffusion. An adaptive anisotropic coefficient, derived from a neuro-fuzzy network, is introduced. The optimal diffusion parameters are estimated using a stochastic optimization technique, which ensures that heavy diffusion performs across the block boundaries and in flat regions while image edges are well preserved. Experimental results have shown that the proposed adaptive diffusion method can significantly suppress coding artifacts and improve the visual quality of the compressed image.