基于光谱特性的高光谱图像压缩方案

根据干涉型高光谱成像仪成像特点,提出了一种针对干涉型光谱仪所获得高光谱图像的基于光谱特性的图像压缩方案。由于光谱信息最终从“点”十涉图像中恢复,因此方案中首先通过高精度匹配技术将原始“像面”干涉图像序列“重组”成“点”干涉图像,然后针对“点”干涉图像序列进行压缩。在重组过程中采取基于光流的亚像素级匹配和基于梯度的三角插值算法,实现了高精度的图像匹配重组;在压缩环节利用“点”干涉图像与光谱信息之间的傅立叶变换关系,提出一种能够很好保持频谱特性的基于一维DCT的压缩算法。实验证明,压缩算法总体性能远高于针对“像而”干涉圈序列的压缩算法．很好地控制了光谱信息的失真。

The Hyper-Spectral Image Compression Scheme Based on Spectral Character

Regarding the characteristics and the imaging mechanism of LASIS(Large aperture static imaging spectrometry),a new compression scheme of interferential hyper-spectral images based on spectral character is proposed.There are two key techniques in the scheme: the first is a new matching algorithm based on optical flow which is applied to implement high precision matching of interferential hype-spectral image;the second is a novel compression algorithm based on 1D-DCT which is proposed to compress "POINT" interferential image directly.According to the features of original interferential image cube produced by LASIS,the image cube is regrouped as "POINT" interferential image sequences firstly.Due to the relationship of Fourier transform between the spectral information and "POINT" interferential image,one dimensional DCT is used to remove the redundancy for each row of "POINT" interferential image.A high efficient entropy coding method combined Run-length with context-based arithmetic coder is realized to compress the quantized DCT coefficients.The experimental results show that the scheme proposed in this paper can compress hyper-spectral images efficiently and it is much better than the method which compresses interferential image directly at the aspect of preserving the spectral information.