离散彩色图像可视模型与快速聚类技术研究

Advanced Color Clustering Algorithm for Discrete Color Image Compression

本文研究了离散彩色图像视觉特征模型，提出了图像可视信息熵、可视质量ＳＮＲ（ＶＰＳＮＲ）、步进聚类等概念；并在该模型基础上，针对离散彩色图像的高效压缩，提出了一种快速聚类算法。算法根据图像ＨＳＶ空间的特性和空域分布特性进行聚类和量化。在保证图像主观视觉质量不变的情况下，使图像的信息数据充分接近可视信息熵。将离散彩色图像视觉特征模型和聚类技术应用于离散彩色图像的压缩，实验表明：系统压缩率可达 ６０～ ３００，与 ＪＰＥＧ相比有明显的提高，提得平均压缩率约为 ＪＰＥＧ的 ６倍。

In this paper, visual model of discrete color image and conception of visual entropy and visual PSNR (VPSNR) are developed. A new clustering algorithm for compression based on this model is also studied and realized. the clustering algorithm detects colors baset on both HSV agglomeration and spatial distribution, it quickly and sufficiently decrease the data of the image to close with visual entropy with high fidelity of the visual feature of original image. Using this theory and algorithm in compression, experiments demonstrate that good R(60-300) which is averagely about 6 times than JPEG are archived.