论文部分内容阅读
针对多波段遥感图象的空间和谱间结构特点，提出了多方式预测的概念：一幅图内的任一象素按照给定的准则在侯选预测函数集中选取其实际采用的预测函数，从而更大程度去除相关；同时利用谱间结构相关，令谱间邻点选用相同的预测方式，使多方式预测导致的附加存储代价大大缩小。提出以极小熵原则作为选取预测函数的理论判据，并将其等效为最小误差的最高频次准则。对ＴＭ图象的实验证明此方法能更有效去除相关，压缩比有较大提高。
According to the spatial and spectral structure characteristics of multi-band remote sensing images, the concept of multi-mode prediction is proposed. Any pixel in a map selects its actual prediction function from a set of candidate prediction functions according to a given criterion. So as to remove the correlation to a greater extent; meanwhile, by using the correlation between spectrum structures, the same prediction method is selected for adjacent points in the spectrum so as to greatly reduce the additional storage cost caused by the multi-mode prediction. The principle of minimum entropy is proposed as the theoretical criterion for selecting a prediction function, which is equivalent to the highest frequency criterion of minimum error. Experiments on TM images prove that this method can remove the correlation more effectively, and the compression ratio is greatly improved.