The Research On Lossless Compression Of Hyper-spectral Images

In the paper, we analyze the characteristic of hyper-spectral images. We compare them with common images, and then discuss the characteristic of sub-images hyper-spectral images through wavelet transform. The experiments show that, compared with common images, hyper-spectral images have insignificant spatial correlation and significant spectral correlation, and through wavelet transform, there are spatial and spectral correlations in the sub-image of hyper-spectral images.Combining the characteristic of hyper-spectral images, and making use of the spatial and spectral correlations within hyper-spectral images, we design a 3-D adaptive predictor which makes the square error minimal, and can use by several bands. Then we use software JPEG-LS to remove the spatial redundancy efficiently. Experiments show that the algorithm works better than JPEG-LS and software WinRAR, and it works a little better than the optimal linear algorithm.What's more, we use lifting scheme 5/3 wavelet to analyze every band of hyper-spectral images. To the same sub-band of different bands, we use our 3-D adaptive predictor and JPEG-LS to compress. The rate of compress is much higher than the anterior algorithm. It possesses great practicality.At last, according to Prim algorithm and the inter-correlation coefficient of hyper-spectral images, we order the sequence of bands anew. Then we use a new adaptive predictor which can use in several bands and JPEG-LS to compress for the newly-ordered bands. Experiments show that the compress rate increases companying with the number of the bands used for predicted.