Reduction of image coding artifacts using spatial structure analysis

This paper proposes a new method to remove ringing artifacts without blurring textures. The method uses LSF (line spread function), color similarity and perceptual similarity to detect coding artifacts in decoded images. Blocks are roughly classified into different types by using LSF values. Perceptual similarities among classified blocks are measured with rules that simulate human visual experience to construct homogenous regions. And color similarities between a pixel and neighboring homogenous blocks are measured to identify ringing artifacts. Non-edge pixels in edge blocks are detected as ringing artifacts if (1) their LSF values are higher than LSF values in nearby flat regions and (2) there are high color similarities between them and flat regions. Experimental results show that our method can preserve image details while removing ringing artifacts and needs no coding parameters. It is superior to MPEG-4 standard method in de-ringing performance, and also moderately reduces blocking artifacts.