Hyper-spectral Image Compression Based On Vector Quantization

With wide application of hyper-spectral remote sensing in the field of nationaleconomy, the imaging spectrometer technology got rapid development. The number ofspectral channels can reach hundreds and spectral resolution becomes higher and higher,and thus the datasets of hyper-spectral remote sensing image increase largely. The hugedatasets have brought great press for data storage and transmission, which restricts thepractical applications of hyper-spectral remote sensing image. Therefore, it is necessary tocompress hyper-spectral remote sensing image by efficient compression techniques.As an efficient image data compression technology, vector quantization hasadvantages of high compression ratio, simple and fast coding and decoding. Vectorquantization has been widely used in the field of hyper-spectral remote sensing imagecompression. This thesis will describe the features of hyper-spectral remote sensingimage and the basic principle of vector quantization. Aiming at overcoming theshortcomings of the existing hyper-spectral remote sensing image compression algorithmbased on vector quantization, this thesis proposes the improved algorithms that based onthe algorithm features and graphics thinking:1. LBG algorithm has the defects of large amount of calculation and poor quality ofimage restoration. Combining with the difference theory, this thesis puts forward ahyper-spectral signal fast difference vector quantization coding method. Throughsimulation experiments, the results show that the improved algorithm is superior to theLBG algorithm in image recovery quality and complexity at the same codebook size.2. The existing hyper-spectral remote sensing image compression algorithms arebased on the Point-to-Line model from the geometrical vector difference ideology, sothey have the disadvantage of large amount of calculation when search the optimal vectorsurface for a vector. Therefore, an improved hyper-spectral remote sensing imagecompression algorithm combining with the characteristics of wavelet transform and basedon the wavelet domain sub-vector Point-to-Line model is put foreword. The experimentalresults show that the amount of calculation is reduced greatly, at the same time theinformation entropy of the image after the quantification is lower.3. A vector quantization in hadamard transform domain is used to the hyper-spectralremote sensing image compression. The calculation complexity of algorithm can be improved. Combining with the features of hadamard transform, this thesis raises ahyper-spectral image compression algorithm with dimension segmentation quantizationon each vector. Through vector segmentation, the algorithm make each part vector tomeet the requirements of the hadamard transform. Experimental results show that theimproved algorithm is superior to the initial algorithms in image recovery quality andcomplexity at the same codebook size.4. On the basis of the hyper-spectral remote sensing image compression algorithmwith dimension segmentation quantization on each vector and multistage vectorquantization, this thesis comes up with an improved algorithm of hyper-spectral remotesensing image compression algorithm. Experimental results show that the improvedalgorithm is superior to the initial algorithms in image recovery quality, complexity andcompression ratio at the same codebook size.