Research On Hyperspectral Remote Sensing Image Compression Algorithm

In the earth-observing system, hyperspectral remote sensing images are important data resource. They have a wide range of application in the national economy and development of society. The hyperspectral images which have three dimensions are very huge. It is hard to transmit and restore. So the hyperspectral remote sensing images must be compressed in applications. The existed image compression standards are for ordinary two-dimension images. There isn't any authoritative compression method for hyperspectral remote sensing images. In this paper, algorithms of hyperspectral remote sensing image compression are proposed.The correlation of hyperspectral remote sensing images is tested at first. The results show: The hyperspectral remote sensing images have much stronger spectral statistical correlation and spectral structure correlation than the RGB color images do. The spatial correlation of hyperspectral remote sensing images is weaker than that of ordinary images. In the compression algorithm design, the central focus is to get rid of the spectral correlation.The NMST (Near Minimum Spanning Tree) algorithm is proposed based on the MST (Minimum Spanning Tree) algorithm. The construction time of MST is too long when the image is large. This is improved in the NMST algorithm. The steps of constructing NMST are: Firstly, construct the derived image which is generated according original image; Secondly, construct minimum spanning tree of derived image; At last, construct spanning tree of original image by adding edges to the minimum spanning tree of derived image. The NMST is taken as prediction tree and is used to remove the correlation of hyperspectral image. Compared with the MST algorithm, the construction speed of NMST is improved more than ten times. The compression ratio of NMST algorithm is decreased less than 5%.The compression algorithm based on bit plane transform is proposed. At first weconstruct a new image transform: bit plane transform. We study on the property of bit plane transform and the application in hyperspectral image compression. For hyperspectral image, different bit plane images have different spectral correlation. The higher bit plane images have strong correlation while the lower bit plane images have weak correlation. The difference is considered in the bit plane transform. Calculation is spent on removing the spectral correlation of higher bit plane images. It improves the calculation efficiency greatly and saves processing time. The calculation complexity of compression algorithm based on bit plane transform is O(N), N is the number of pixels in the image. The compression ratio of algorithm is more than 1.9. The compression algorithm based on bit plane transform can be realized by parallel computing model. The calculation for high bit data and low bit data are independent. There isn't communication and synchronization in the whole process. The calculation of the bit plane transform is operation of bit mainly. The algorithm can be realized by hardware.