Application of JND visual model to SPHIT image coding and performance evaluation

The wavelet based SPIHT algorithm has recently attracted significant attentions from researchers in image coding. SPIHT maximizes PSNR (or minimize MSE) for each bit it transmitted, however, it is well known that PSNR is not a good indicator of the perceptual image fidelity. Human visual perception is more sensitive to distortions in lower frequency components than distortions in the higher frequency components. In this paper, we enhance SPIHT by introducing a process called JND_SQ, between wavelet transform and SPIHT. JND_SQ allows SPIHT to maximize perceptual fidelity instead of PSNR for each bit transmitted. We proposed a perceptual fidelity criterion called JND_PSNR based on the SY JND model, the new criterion is more consistent with human perception than PSNR. Experimental results show that JND_SPIHT outperforms SPIHT in terms of JND_PSNR. The JND_SQ process results in a large amount of zeros, which can be further utilized to improve the speed of SPIHT. Finally, JND_SQ can be turned off if the original SPIHT is desired.