Research On Efficient Compression For Hyperspectral Images

Interferential hyperspectral imaging technologies are considered as the practical imaging spectrometers in the current remote sensing. An interferential spectrometer, which can obtain plenty of both spatial and spectral information of a target, is widely used in environmental monitoring, geology, meteorology, military and other fields. As a three-dimensional image, the mass of data lead to the problems on transmission on the limited channel bandwidth and the storage on satellite. Therefore, it is necessary to design efficient compression methods according to the features of interferential hyperspectral image.Based on the aerospace applications, the interferential hyperspectral image compression technologies are fully studied in this dissertation from two aspects: one is in lossless compression aspect to study the improvement and combination of prediction and Vector Quantization (VQ) to design high performance on compression for interferential hyperspectral image; the other is in lossy compression aspect to study three-dimensional wavelet transform and coding approaches which effectively maintain the spectral and spatial information of the reconstructed images.The main contributions are as follows:1. Fast dual-direction prediction lossless compression is proposed for united modulation interferential hyperspectral spectrometer data-Large Aperture Static Imaging Spectrometer (LASIS) data. This method consistes of two modules, dual-direction prediction and entropy coding. Dual-direction prediction can eliminate both spatial redundancy and spectral redundancy in LASIS data. Meanwhile an optimization is proposed for dual-direction prediction to enhance the speed of operation. Range coder is adopted to speed coding processing. Fast dual-direction prediction lossless compression method suitable for practical application of space-board compression outperforms many current international popular lossless compression methods both on compression ratio and speed.2. A point to line model for the expansion of VQ algorithm is proposed. Meanwhile a lossless compression scheme using this expansion of VQ is carried out on interferential hyperspectral image. Based on Linde-Buzo-Gray (LBG) algorithm, the expansion of VQ makes use of perpendicular connecting the point to the line to represent vector for the quantization. Combined with dual-direction prediction method, the lossless compression method of dual-direction prediction expansion VQ can remarkably improve the compression performance of interferential hyperspectral image.3. A three-dimensional (3D) orientation prediction filter in lifting structure is proposed for LASIS data. The novelty of this 3D-wavelet is that it combines directional prediction into 3D-wavelet. In this way, the prediction selects the most correlative pixel rather than pixels limited on the prediction direction. The experimental results show that this 3D orientation prediction wavelet improves the performance of wavelet obviously, particularly on the LASIS image with quite severe direction.4. Weighted Rate-Distortion Optimization for Set Partitioning in Hierarchical Trees (SPIHT) is proposed for interferential hyperspectral image lossy compression. According to the traits of images in sptial domain and Fourier domain, it is testified that pixels on the large optical path difference are important for the shape of spectrum. So a rate-distortion optimizer for the SPIHT is designed to give an adaptive lifting of the rate-distortion slopes, which not only makes up the deficiency of bit allocation in SPIHT, but also protects the spectral information efficiently. The experimental results show that the proposed algorithm achieves improved performance over the conventional algorithms both in sptial domain and Fourier domain.