Research On Hyperspectral Image Compression Algorithm

Since 60's, the technology of remote sensing has developed fast. People use hyperspectral images in many fields such as exploiting resources, monitoring environment, military scouting and so on. As the development of hyperspectral imager, the hyperspectral images data increase largely, and bring huge pressure on data transmission and storage. Because the remote sensing information is precious and we expect high exactitude of resumed information, so we give more concern on lossless or nearly lossy compression. In this paper, we have done some work on lossless hyperspectral images compression as follows:First of all, we analyze the characteristic of hyperspectral images, and compare it with common images. The experiment shows that, compared with common images, hyperspectral images have abundant texture and insignificant spatial correlation. And result shows that their spectral correlation is significant. In conclusion, when we compress hyperspectral images, we should focus on remove their spectral correlation.Secondly, hyperspectral images are hard to compress because of their abundant details, complicated texture and insignificant special correlation. Making use of the significant spectral correlation within the hyperspectral images, we propose an optimal linear predictor which makes the square error minimal. Then we use lifting scheme and SPIHT algorithm to remove the spatial redundancy efficiently. Experiments show that the method works better than 3D-SPIHT algorithm and software WINRAR.What's more, we design a 3D prediction compression scheme. The scheme is based on our optimal linear predictor and we use JPEG-LS lossless compression algorithm to compress the residual images. The scheme costs less time in computing, but works much better than JPEG-LS algorithm and software WINRAR. It possesses great practicality.At last, according to the significant spectral correlation of structure within the hyperspectral images, we propose a hyperspectral image lossless compression algorithm based on classification and prediction. Each band of hyperspectral image has the same physical structure, so we classification the first band, and design an optimal linear predictor for each class to make the mean prediction square error minimal, and then we use JPEG-LS algorithm to remove the spatial redundancy. Experiments show that compression ratio of this algorithm increases by 0.02 more.