Implementation Of CCSDS Hyperspetral Image Lossless Compression Based On Tilera Processor

Hyperspectral image is widely used in many aspects, such as agriculture, land planning, environmental monitoring, urban planning and resource explorating. The special resolution and spectral resolution of hyperspectral images are getting higher and higher, and the increasing amount of image data leads great pressure to transferring and storaging hyperspectral images, so it's necessary to compress hyperspectral images. Many fields have high quality requirement on hyperspectral images, which do not allow any image distortion, so compressing hyperspectral images losslessly is the best choice. Among the existing hyperspectral image compression algorithms, CCSDS 123.0-B-1 algorithm is an international standard of multispectral & hyperspectral image compression, which is developed by CCSDS group.Both compression ratio and compression rate are major indicator of the performance of image compression algorithm. Lossless compression does not allow information distortion, so it's hard to increase compression ratio substantially, but we can improve the whole performance of the algorithm by increasing compression rate. Lossless image compression algorithm used to be realized based on single-core processor. After years of development, the development of single-core processor has come to an end. It's hard to significantly improve the performance just by improving single-core processor. Multi-core processor can improve the computational performance of processor greatly. Therefore using multi-core processor to compress image can increase compression rate significantly.In order to solve the above problems, this paper studies CCSDS 123.0-B-1 algorithm. The main innovative points are as follow:(1) This paper improves the prediction part of the algorithm, and then proposes a prediction algorithm based on detecting edge, so as to increase compression ratio.(2) This paper realizes the improved CCSDS 123.0-B-1 algorithm based on TILE-Gx36 multi-core processor which is produced by Tilera company, so as to increase compression rate.The prediction algorithm based on detecting edge proposed at this paper makes use of the edge information of image and detects the direction of edge according to the dynamically refreshed threshold. According to the direction of edge, the improved algorithm assigns different weight to neighbor pixels, so as to predict more accurately. The simulation result shows that the improved algorithm improves compression ratio by about 1.5% to 11.2% and doesn't reduce compression rate obviously.The concurrent program this paper realized mainly includes two modular, hyperspectral image slicing modular and memory management modular. In this concurrent program, because of the serial steps of the algorithm, hyperspectral image is sliced into small hyperspectral images and each small image is encoded in different cores at the same time. Cutting image into small ones would cause information loss, but simulation result shows that the amount of information lost is acceptable. Frequently applying and releasing memory would cause performance loss, so this concurrent program applies all the required memory and manages them with four chain tables. Simulation result shows that the concurrent program greatly improves compression rate of the algorithm by about 13 times.