Research On Hyperspectral Image Compression Based On Vector Quantization

Hyperspectral images are obtained simultaneously on multiple narrowband spectral bands by observing the same ground.It reflects the characteristics of the observed object's electromagnetic radiation in various narrow spectral bands and contains much information of the observed object.Due to the large amount of hyperspectral remote sensing image data,it is difficult to store and transmit.Using vector quantization technology to compress hyperspectral images can achieve high compression ratio,while retaining more original information.However,the existing hyperspectral image vector quantization compression algorithms have high complexity and large distortion of the compressed image.Therefore,in order to settle these issues,based on the analysis of hyperspectral image data,this paper proposes two improved hyperspectral image vector quantization compression algorithms,which significantly improve the performance of vector quantization of hyperspectral image.The specific achievements are as follows:Firstly,according to the improved effect of anomaly detection on hyperspectral image compression,a fast vector quantization compression algorithm for hyperspectral image based on anomaly detection is proposed.The algorithm improves the original anomaly detection algorithm using PCA and finishes vector quantization compression using LBG algorithm.Experiments results show that the proposed algorithm not only can improve the compression quality but also reduce the amount of computation to achieve fast hyperspectral image compression.Secondly,according to the idea of the improvement of impression after classification,a vector quantization compression method for hyperspectral image based on band clustering and image classification is proposed.Through the band clustering + PCA and image classification,the vector quantization codebook search scope and the search operation are reduced.Experiments results show that the algorithm can greatly reduce the computational complexity and ensure the fast compression of hyperspectral images under the premise that the quality of image restoration is basically unchanged.