摘要
本文以小波作为编码框架,通过对人工合成图像的主观实验,得出小波变换域中各不同频带的视觉误差阈值函数,进而提出了一个新的量化编码方案。该方案基于小波域非均匀零树结构量化和高频低能系数的视觉非均匀量化。对实际图像的感知熵和几乎透明的情形作了研究,与JPEG标准也作了比较。实验表明该方案较充分、自适应地考虑人眼视觉冗余,具有稳健、实际可行的特点。对自然图像,在十倍压缩比左右可有视觉无失真质量。
In this paper, perceptual error threhhold (PET) functions in different wavelet subbands are obtained through subjective tests on artificial patterns. A new and generic two-stage coding scheme is thus proposed, based on the nonuniform zero-trees of wavelet coefficients and the visually nonuniform quantization of small coefficients in higher subbands. Researches have been done for the visual transparent and nearly transparent cases, in comparisons with JPEG standard. Experimental results showed the feasibility and robustness of the scheme. Transparent quality can be obtained around 10 compression ratios for natural images.
基金
国家自然科学基金
关键词
感知编码
感知熵
小波变换
零树
图像编码
Perceptual coding, Percetual entropy, Wavelet transform, Zero-tree
