Image coding using parametric texture synthesis

Abstract

Visual textures like grass, water etc. consist of dense and random variations in contrast that are perceptually indistinguishable by a human eye. Such textures are costly to encode using image and video codecs. For example, in the state-of-the-art compression standard High Efficiency Video Coding (HEVC), detailed textures typically show relatively strong blurring artifacts at low rates (high QPs). Texture synthesis is a process whereby one can obtain a reconstruction of a visually equivalent texture with decent visual quality, given a set of parameters. In this paper, texture synthesis is used as a tool in combination with HEVC, exploiting Human Visual Perception (HVP) properties by creating an artificial textured content using model parameters at the decoder side. A novel scheme for compression (prediction and quantization) of parameters for complex wavelet based texture synthesis is introduced. The compressed parameters are sufficient to synthesize high quality texture content at the decoder side. Simulation results have shown, that with same rates, both the subjective and the objective quality is enhanced, compared to HEVC.