基于波段分组的高光谱图像无损压缩

高光谱图像海量数据给存储和传输带来极大困难,必须对其进行有效压缩。针对高光谱图像不同频谱波段间相关性不同的特点,提出一种基于波段分组的高光谱图像无损压缩算法。为了降低波段排序算法的计算量,根据相邻波段相关性大小预先进行分组,采用最佳后向排序算法对各组波段进行重新排序。在当前波段和参考波段中选取具有相同空间位置的邻域结构,在最小二乘准则下,利用邻域像素对当前预测像素进行最优谱间预测。参考波段和预测残差数据进行JPEG-LS压缩。对OMIS-I型高光谱图像进行实验的结果表明,与基于多波段预测算法相比,该算法可进一步降低压缩后的平均比特率。

Lossless Compression for Hyperspectral Images Based on Waveband Grouping

The data size of hyperspectral images is too large for storage or transmission, so it is necessary to compress hy- perspectral images efficiently. Since the characteristic of spectral correlation differs between different bands, a new lossless compression algorithm for hyperspectral images based on spectral band grouping is proposed. The spectral band grouping to divide spectral bands is introduced into groups according to the correlation coefficients. The optimal backward reordering algo- rithm is adopted to reorder spectral bands of each group efficiently. The neighborhood structure with same space location is selected in the current band and the reference band. The neighborhood pixels are utilized to predict the current prediction pixel within the optimal spectra based on least squares rule. Finally, JPEG-I.S standard compression of reference bands and predic- tive residual data is carried out losslessly. Experimental results show that the proposed algorithm can further improve compression performance in comparison with previous methods based on multi-bands prediction.