Performance improvement of wavelet-based scalable video coder based on pre-decoder side rate-distortion optimization

Abstract

In wavelet-based scalable video coding, the encoder generates one large bit-stream and the pre-decoder can truncate it to have an arbitrary size by using the embedding principle. Most conventional rate control algorithms utilize some information available only at the encoder, thus, it cannot be applied directly in the scalable video coders since the actual bit allocation should be performed after the encoding phase. We modify the rate-distortion (R-D) function and propose a new rate control algorithm based on the modified R-D function, which uses only the pre-decoder side information. In addition, since large PSNR variation results in noticeable visual artifacts at low PSNR frames, we propose a novel distortion flattening technique to reduce the variance of PSNR curve and to improve the performance of low PSNR frames. Experimental results have shown that the both proposed methods improved the average PSNR, and especially, the proposed distortion flattening technique makes the visual quality more consistent.