Wavelet-based color image compression: Exploiting the contrast sensitivity function
Wavelet-based color image compression: Exploiting the contrast sensitivity function
复制标题
DOI：
10.1109/tip.2002.807358
复制
发表时间：
2003-01-01
影响因子：
10.6
通讯作者：
Kunt, M
中科院分区：
文献类型：
--
作者：
Nadenau, MJ;Reichel, J;Kunt, M
图像压缩技术的视觉效率直接取决于其保留的视觉重要信息量。 “视觉上重要”是指人类观察者最敏感的信息。总体灵敏度取决于对比度、颜色、空间频率等方面。一个重要的方面是对比敏感度与空间频率之间的反比关系。这是通过对比敏感度函数（CSF）来描述的。在压缩算法中，可以利用 CSF 来调节量化步长，以最大程度地减少压缩伪影的可见性。现有的基于小波的图像压缩的 CSF 实现对大范围的空间频率使用相同的量化步长。这是 CSF 的粗略近似。本文提出了两种新技术，可以以更高的精度实现 CSF，甚至可以适应分解子带内空间频率的局部变化。这些方法可用于亮度图像和彩色图像。对于颜色感知，三个不同的 CSF 描述了灵敏度。每个色带的实现技术都是相同的。新技术被实现到 JPEG2000 压缩标准中，并与传统的 CSF 方案进行比较。所提出的技术在视觉上比之前发布的方法更有效。然而，本文的重点是如何以精确和局部自适应的方式实现 CSF，而不是这些技术的优越性能。
The visual efficiency of an image compression technique depends directly on the amount of visually significant information it retains. By "visually significant" we mean information to which a human observer is most sensitive. The overall sensitivity depends on aspects such as contrast, color, spatial frequency, and so forth. One important aspect is the inverse relationship between contrast sensitivity and spatial frequency. This is described by the contrast sensitivity function (CSF). In compression algorithms the CSF can be exploited to regulate the quantization step-size to minimize the visibility of compression artifacts. Existing CSF implementations for wavelet-based image compression use the same quantization step-size for a large range of spatial frequencies. This is a coarse approximation of the CSF.This paper presents two new techniques that implement the CSF at significantly higher precision, adapting even to local variations of the spatial frequencies within a decomposition subband. The approaches can be used for luminance as well as color images. For color perception three different CSFs describe the sensitivity. The implementation technique is the same for each color band.Implemented into the JPEG2000 compression standard, the new techniques are compared to conventional CSF-schemes. The proposed techniques turn out to be visually more efficient than previously published methods. However, the emphasis of this paper is on how the CSF can be implemented in a precise and locally adaptive way, and not on the superior performance of these techniques.