The Research Of Cube Of Interest Protection Based Hyperspectral Remote Sensing Image Compression

The relatively high spectral resolution is very important for the application of hyperspectral remote sensing images. However, the massive data of hyperspectral images leads to a huge pressure to data storage and transmission. Due to the special characteristics of hyperspectral remote sensing images, there are significantly difference between hyper spectral remote sensing images and general two-dimensional images. Therefore, it is an important topic to research the compression of hyperspectral remote sensing images according to their special characteristics.Comparing with general two-dimensional static images, the spatial correlation and spectral correlation are analysed in this paper. The results show that the spatial correlation of hyperspectral images are weaker than that of two-dimensional static images, and the spectral correlation of hyperspectral images is stronger than its spatial correlation. The superiority of FastICA algorithm mainly relies on the dimension reduction of initial data (provided by Minimum Noise Fraction (MNF) Rotation) that reduces the adverse effect of noise to object detection.The interest concept, namely the region of interest (ROI) in the the band of interest (BOI), is put forward in this paper. According to the application of ground object classification, the selection method of the optimum band is studied. According to the application of object extraction, the wavelet-based band selection method is also investigated. Moreover, the effects of the two band selection methods are verified through the simulation experiment. During the ROI extraction, the FastICA algorithm is studied in this paper. Experiments validate that FastICA performs better than other state-of-art methods in terms of highlighting objects and suppressing background.Based on wavelet Contourlet transformation and COI protection, the compressing method of hyperspectral images is studied proposed in this paper. On one hand, when conduct the low bit rate compression for images with rich texture information using 2D discrete separable wavelet transform, the ringing phenomenon will occur in image edges. On the other hand, there remains 4/3 degree of redundancy when just utilizing Contourlet transformation for image compression. Therefore, the compression method combining these two methods is studied in this paper (wavelet and Contourlet) and achieves their complementary advantages. The content of COI protection is more accurate than those of simple ROI protection or BOI protection. The information amount of COI significantly cuts down compared with the simple ROI protection or BOI protection. Therefore, the reconstructions of ROI and background with the same compression rate both obtain superior image quality. Extending Maxshift algorithm from two-dimensional images to three-dimensional images raises the wavelet coefficient of ROI in BOI. Experimental results show that the peak signal-to-noise ratio of the proposed method beyonds 40dB which increases by 3.16dB compared with that of 3D-SPIHT algorithm when the compression ratio of 8:1. Due to the combination of Contourlet transformation, the textures and the edges of reconstructed images appear more distinct and clearly than those of original images.The reconstructed images used to classification application almost unaffected.