A scalable coding based on structure-preserving filter and saliency-ordered transmission

Abstract

This paper presents a new scalable coding (SC) algorithm considering the properties of human visual system (HVS), especially the importance of structures and saliencies for visual perception. The proposed method first finds the salient regions in an image and then applies structure preserving filter to the regions excluding the saliencies. This image is encoded with a pre-specified quantization parameter (QP), which is the base layer in our coding system. Unlike the base layer image in the conventional methods, our base layer keeps the important regions (usually foreground) and also structures in the background while suppressing noise and fine textures. In addition, the enhancement layer is sent in the order of saliencies, i.e., the residual signals in the salient regions are given the priority in the transmission of enhancement layer. The algorithm is implemented based on the HEVC intra coding, and the experimental results show that the proposed method yields higher PSNR in salient regions than the conventional SHVC (scalable HEVC). Also, the overall perceptual quality is improved because the quality in the salient region is improved and important structures in the background region are kept, while the smeared noise and fine texture in the background region do not much affect the perception.