Perceptual Quantization for Predictive Coding of Images

Kyeong Ho Yang, A. Farid Faryar, Wenwu Zhu

Abstract: In this paper, we propose a new perceptual quantization method for predictive coding of images, where the quantization step size is adaptively changed according to the perceptibility of the distortion incurred to the current pixel. We measure the perceptibility using only the previously reconstructed samples which are also available at the decoder, so that any quantization-related overhead information does not need to be transmitted to the decoder. The proposed method can also provide the lossless concatenated coding property when applied to multigeneration of images since the same information on the perceptibility is available at the second time encoding. experiments show that the proposed perceptual quantization method provides much better subjective picture quality than the non-perceptual one when applied to the JPEG-LS standard.