JPEG、JPEG2000图像压缩与色域变化的关联性分析

在传输带宽或存储空间有限时，利用图像压缩技术缩减冗余图像信息，可以高效地传递、储存数据。但是由于有损压缩的不可逆性，图像在高压缩率下会出现很多颜色失真，颜色信息的改变也会影响到压缩图像的色域属性。本文对使用JPEG和JPEG2000压缩图像的压缩比与色域变化间的关系进行了研究。为了探究两者之间的关系，同时考虑到其在CIELAB颜色空间中的分布均匀性，以Macbeth Color Checker中的18种颜色样本为代表。基于代表色的颜色信息，按照4种颜色属性将12幅自然彩色图像分为2组，检验压缩图像的压缩比与色域的关系，发现色域保真性与不同色域中颜色数量和压缩比有关。最后，使用最小二乘法得到压缩比与色域变化量间的关系。对每组图像，通过二次方程绘制拟合曲线，得到原图形数据、压缩图像色域和对应近似值间的最小误差。

Analysis of Relationship between Image Compression and Gamut Variation： JPEG and JPEG2000

Image compression schemes facilitate the transmission and storage of data in an efficient form by reducing redundant image data in the case of limited transmission bandwidth or storage space. However, due to the irreversible nature of lossy compression, these schemes also introduce various types of color artifacts into the reconstructed image at a high compression ratio. Consequently, a change in color information affects the gamut characteristic of the reconstructed image. Accordingly, this article investigates the relationship between the compression ratio and the gamut variation for a reconstructed image using JPEG and JPEG2000. To analyze the relationship between the compression ratio and the gamut variation, 18 color samples from the Macbeth ColorChecker are distribution in CIELAB color space. Based on the images, classified into two groups depending on fo initially used as representative colors for all colors due color information shift for representative color samples, ur color attributes, are used to investigate the relationship to their uniform 12 natural color between the level of compression and the variation in the gamut area for reconstructed images. After determining the gamut areas for the decompressed images in relation to the compression ratio, gamut fidelity is obtained using the ratio of unique colors relative to the gamut area. Finally, the optimal least square method is applied to obtain fitted curves as an equation minimizing the error between the real data, the gamut area for the decompressed images, and the corresponding approximated values.