一种适于高光谱图像压缩的相关系数矩阵近似计算算法

An approximate algorithm for correlation matrix computation in hyperspectral image compression

为充分利用高光谱图像的强谱间相关性,许多学者提出了基于相关系数矩阵的波段预处理的图像压缩算法,得到很好的压缩性能。但是由于相关系数矩阵计算量过大,严重制约了波段预处理方法的实时性。文中提出一种相关系数矩阵近似计算算法,通过对高光谱图像进行空间域采样,以很小的计算量求得近似的相关系数矩阵。用真实高光谱图像数据的实验结果表明,该算法在几乎不影响压缩性能的前提下,能大幅降低计算量,使得波段预处理方法的实时实现成为可能。

For exploiting the strong spectral correlation of hyperspectral images, many correlation matrix-based band preprocessing methods is proposed to improve the compression efficiency. However, the real-time performance of band preprocessing method is restricted by the high computational burden of correlation matrix. Since an approximation algorithm that sample in spatial domain of hyperspectral images is adopted in this article, an approximation correlation matrix can be obtained with much lower calculation. Experimental results show that, the proposed algorithm hardly reduces the compression performance and sufficiently reduce the computational cost, which can improve the real-time performance of band preprocessing method.