基于区域的多光谱图像无损压缩

针对多光谱图像的冗余特点,提出一种多方式预测的多光谱图像无损压缩方法.多光谱遥感图像每一波段的图像本身存在空间冗余.不同波段图像之间存在谱间冗余.由于不同波段的成像对象是同一地域,具有相同的物理结构,虽然同一空间位置不同波段的像素值相差很大,但其与邻域的关系却是很相似的,同类地物在图像中占据一定区域,在不同的波段具有相似的纹理.因此,首先选择基于区域生长的分-合算法将多光谱图像分割成若干块互不重叠的均匀子区域;又由于各个子区域的纹理不同,因此对不同区域采用不同的预测方式,从而最大程度去除冗余;同时利用谱间结构相关,令各波段相同区域选用相同的预测方式,共享预测方式图可使多方式预测导致的附加存储代价大大缩小.实验结果表明了该方法的有效性.

Region-based lossless compression for multispectral image

A new technique in lossless compression of remotely sensed imagery was proposed.Due to the imaging principle of multispectral imagery (MSI),similarity exists between the pixels on different bands with the same spatial site and their neighborhoods,thus leading to structure correlation.The method introduced manages to reduce such correlation based on multi-schemes predictive coding.Since the gray values and textures of each region are quite different,the original image was divided into several sub-regions,then compressed with different methods,in order to obtain a better compression ratio.On the other hand,the same segmentation was done and the same prediction mode was used for each band within a sub-area of the same site.So the extra cost of the prediction map will be shared by all the bands and reduced.The experiment results show that the method is favorably comparative with other compression techniques on several Landsat-TM images.