目标保真的背景影像高倍率压缩

传统的分类压缩方法由于对背景采取一致性的有损压缩，在高压缩后不能保留细节，因此在图像重建时背景不能提供很多的信息，整图重建的效果取决于目标本身的分割质量。本论文在将目标提取后，再对背景进行细分类，保留与目标关系密切的部分背景，并把这部分区域与已分割出的目标划为同一类，其余背景划为一类，进行分类别压缩。这种方法允许分割的目标近无损，能增强目标的可理解性。

Features Cluster for Remote Sensing Image Compression

Traditional classified compression methods have some deficiency for remote sensing image. Due to background is highly compressed, image reconstruction depends on segmentation quality for objects at large intent. But in fact segmentation process often leads to lossy objects. This paper improves it by new methods. After the objects is firstly segmented, then the further cluster is carried on for keeping some backgrounds in close relation with objects which are combined with the segmented objects, while other backgrounds are taken for another class and this region can be compressed at a higher compression ratio. The experiment shows the method above allows little lossy segmentation and can strengthen intelligibility of objects.