通过对不同小波系数加权优化量化步长的方式提出了基于视觉特性的零树编码算法。视觉权值通过刚能鉴别出的失真阈值JND计算出来。而JND的计算则通过人眼视觉系统的频率敏感度、亮度敏感度、图像内容的对比掩盖效应以及边缘掩盖效应4个特性来计算。实验证明,在相同主观质量的情况下,提出的算法优于原始的EZW/SPIHT算法,可以取得更高的压缩比。
In this paper we propose a perceptually-tuned embedded zerotree image coder that introduces a perceptual weighting to the wavelet transform coefficients prior to EZW encoding, The perceptual weighting for all subbands are computed based on the just notice- able distortion (JND) thresholds, And the JND thresholds are computed based on the four different properties of the human visual system that have been studied in the context of image coding: frequency sensitivity, luminance sensitivity, contrast masking, and edge masking. The experiment shows that perceptually-tuned coder produce a lower bit-rate than SPIHT, and has the same perceptual image quality.