Predicted-based Lossless Compression Of Hyperspectral Images And Aurora Spectral Images

Since the 1980 s, the hyperspectral remote sensing technology has been developed rapidly. It combines spectral information of determining features with two-dimensional images of reacting space information, which forms a kind of 3 dimensional data. As hyperspectral technology developing in the direction of high spatial resolution and high spectral resolution, the data that acquired by imaging spectrometer increases exponentially. In view of high cost of obtaining, transmitting and storing hyperspectral data, it is very important to design efficient lossless compression algorithm.This paper introduces hyperspectral remote sensing technology and the research status and significance of hyperspectral images and aurora spectral images, proposes a C-DPCM algorithm with adaptive removal of local spectral outliers after analyzing the feature of C-DPCM algorithm. The results of experiment show the compression effect of C-DPCM algorithm with adaptive removal of local spectral outliers is better than that of C-DPCM algorithm. Then, this paper researches new aurora spectral data, presents two-dimensional JPEG-LS and three- dimensional JPEG-LS respectively. At last, put forward a lossless compression algorithm for aurora spectral data using online regression prediction. The main results of the thesis are as follows:Firstly, this paper proposes a C-DPCM algorithm with adaptive removal of local spectral outliers. This algorithm uses two times regression training loaded the function of adaptive removal of local spectral outliers to the C-DPCM algorithm. When performance the first regression training, it uses the prediction coefficients of all spectral lines to predict all image pixels, then compute the residual values. After the first training, we rank the residual values in the order of big to small, and the spectra lines are disconnected adaptively to two kinds in the place where the difference of the adjacent distances is greatest, and the spectra in the cluster with larger distances are local spectral outliers. When executing the second training, we use the rest of spectral lines except from local spectral outliers. The contrast experimental results indicate this method is better than other algorithm in time and compression effect.Secondly, this paper raises a lossless compression algorithm for aurora spectral data based on JPEG-LS algorithm in chapter four. Because a frame aurora spectral image can be seen a two-dimension image, it can uses the core algorithm of JPEG-LS, LOCO-I algorithm, on a frame aurora spectral image, and the compression effect of JPEG-LS is much better than RAR compression and ZIP compression. Next, in view of continuous aurora spectral image as a kind of time series, it is necessary that there is temporal correlation between frames. So we present three-dimensional JPEG- LS inter frame prediction method. We use the inter frame correlation into predictor by updating LOCO-I algorithm. The experimental results show three-dimensional prediction is better.Thirdly, this paper proposes a lossless compression algorithm for aurora spectral data using online regression prediction. This method avoids transmitting the prediction coefficients to decoder, because it adopts least squares algorithm to finish online prediction. The idea of recursive regression is that it uses have been encoded(decoded) pixels to training predictor, and then predict unknown pixels and finish predicting the whole image by constantly update to the training samples participating in the training. After coding, only the residuals are needed to transmit. The experimental results indicate that recursive regression method can improve the effect of compression.The work is supported by national natural science foundation of china(No.61377011) and marine public scientific research projects, the oceanic administration of the people’s republic of china(No.201005017).