Applying Wavelet JND Model to JPEG2000

D.-F. Shen and R.-C. Jian (Taiwan)

Keywords

JND_PSNR, JND_SQ, JND_JPEG2000, JPEG2000, SY JND model.

Abstract

In this paper, A scalar quantization process, which is based on human visual JND (Just Noticeable Difference) property, called JND_SQ is introduced between the DWT and Quantizer in JPEG2000 standard. The proposed JND_SQ pre-quantizes coefficients of different wavelet sub-bands with different step sizes, according to the importance of the sub-band to human perception. In order to prove the effectiveness of the JND_SQ, we introduce two perceptual image quality metrics namely the objective JND_PSNR (intended to replace PSNR) and the subjective VLLCVD (Visually LossLess Critical Viewing Distance) for evaluation of perceived image quality. Experimental results show that the proposed JND_JPEG2000 outperforms JPEG2000 standard not only in bit rate-perceptual quality sense, but also significant reduction in computations; this is because a great amount of coefficients are quantized to zeros by the JND_SQ process.

Important Links:



Go Back