基于分类K—L变换的多波段遥感图像近无损压缩方法

去除空间和谱间相关性是多波段遥感图像压缩中的重要环节，为了得到更好的去相关效果，将矢量量化方法引入多波段遥感图像压缩中，以去除对应同一地物的波段矢量间的相关性，再通过分类K－L变换去除量化误差图像的变间相关性，对K－L变换后的特征图像采用预测树的方法进一步去除谱间结构相关性和空间相关性，实验结果表明，该方法可以取得很好的压缩效果。

Near-Lossless Compression of Multispectral Remote Sensing Image Based on Classified K-L Transform

The spatial and spectral decorrelation are important steps in the compression of multispectral remote sensing image. To obtain better decorrelation effect, in this paper, the vector quantization is employed into the compression of multispectral remote sensing image in order to decorrelate the spectral vectors corresponding to the same objects. Then the classified K_L transform is used to reduce the spectral correlation of quantization error image. Finally, the prediction tree is adopted to reduce the spectral correlation of structure and the spatial correlation of the eigenimages. The experimemtal results show that satisfactory compression effect, has been achieved using the methods introduced in this paper.