Research On The Compression Of Multispectral Remote Sensing Image

Multispectral remote sensing image is acquired by observing the same object (area or target) in multiple narrow wavelength slices at the same time and reveals the reflection, transmission, or radiation features of the observed object in multiple spectral bands. With more information about the observed object contained, multispectral remote sensing image has been widely used in a large number of applications, such as aerospace, mineral prospecting, environment monitoring, lunar exploration, etc. Naturally, compression of multispectral image has become one of the most important research tasks in the discipline of signal and information processing and attracted more and more attentions. As multispectral image can be viewed as a sequence of still gray images with spectral correlations among them, the compression technologies of both still gray image and multispectral image are systemically and deeply investigated in this dissertation. The research work and the main results are as follows:(1) The Dilation-Run algorithm based on wavelet transform for the compression of still gray image Based on the analysis of the characteristics of wavelet transform coefficients, the new wavelet image coder, dilation-run algorithm, is proposed according to the properties that most wavelet coefficients are insignificant and the significant ones tend to cluster. The algorithm integrates the morphological dilation operation and the run-length coding, where the morphological dilation is used to search and encode the clusters of significant coefficients and an efficient run-length coding method is used to encode most of the insignificant coefficients between two clusters. The experiment results show that the new coder outperforms the classic zerotree coder SPIHT and is competitive with two famous morphology coders MRWD and SLCCA. For images with strong clustering feature in wavelet field, the new coder outperforms both the morphology coders above.(2) The Group Based Conditional Entropy Coding of Block Transform Coefficients (GCECB) algorithm for the compression of still gray image Lapped transform, instead of 8×8 block DCT, is used in order to avoid blocking artifacts in the restored images. Based on the analysis of the characteristics of block transform coefficients, the new image coder, GCECB, is proposed. The algorithm encodes the coefficients by 2×2 groups, which mainly makes use of the property that the magnitudes of 2×2 groups decrease more obviously than those of separate coefficients in an 8×8 block. In the algorithm, the output symbols are conditional entropy codedaccording to the intra-block and inter-block correlations of the coefficients. Several methods are also adopted to reduce the complexity of the algorithm. The experiment results show that the new coder outperforms the wavelet coders evidently for images with a large amount of regular textures. For other images, the performance of the new coder is comparable to that of SPIHT with arithmetic coding and obviously exceeds that of SPIHT with binary encoding.(3) Wavelet Coding of Multispectral Image (WCMSI) algorithm for the compression of multispectral image with low spectral resolution After analyzing the properties of multispectral data and the characteristics of its spatial wavelet transform coefficients, the rules about the inter-band structural correlation of the spatial wavelet coefficients are given. On this basis, the coder WCMSI is proposed. The algorithm is a bit-plane coder, whose key feature is the conditional entropy coder designed according to the inter-band structural correlation and the intra-band correlation of the coefficients. Two modes of the algorithm, one without spectral transform and the other one with spectral KLT, are both given. The experiment results show that the new coder outperforms the wavelet coders without the consideration of inter-band structural correlation evidently. For multispectral image with weak inter-band statistical correlation, the superiority of the algorithm is more evident.(4) 3-D Wavelet Coding of Hyperspectral Image (3DWCHSI) algorithm for the compression of hyperspectral image with high spectral resolution Theproperties of hyperspectral data are analyzed. The spectral and spatial correlations of its 3-D wavelet transform coefficients are investigated. On this basis, the new coder 3DWCHSI is proposed. The algorithm is a bit-plane coder, whose key feature is the conditional entropy coder designed according to the spectral and spatial correlations of the 3-D wavelet coefficients. The experiment results show that the new coder outperforms the classic 3-D wavelet coder 3DSPIHT evidently. The influence of segment compressing, which is the main method to reduce complexity, are also tested and discussed.(5) Error Compensated Prediction Tree (ECPT) algorithm for the lossless compression of multispectral image The new lossless coder is proposed according to the inter-band structural and statistical correlations of multispectral data. The algorithm makes use of both the spectral correlations to construct predictor, so the errors created by the original prediction tree algorithm are compensated. For multispectral image with low spectral resolution, an adaptive ECPT algorithm is given. With the consideration of the high computational complexity of the adaptive algorithm,a simplified algorithm based on local stability of image is also designed. The experiment results show that the algorithm outperforms the original prediction tree one. The adaptive ECPT achieves better results than ECPT for multispectral image with weak inter-band statistical correlation, while its complexity after simplification is comparable to that of the non-adaptive algorithm.6. Region-of-interest compression of multispectral image For the situation that users may only have interests in some special region in a multispectral image, a region of interest (ROI) coding algorithm is designed. The algorithm first encodes the shape of ROI by a context-based arithmetic coding method. Then inter-band prediction and intra-band integer wavelet transform are used to remove the spectral and spatial redundancy separately, so that the algorithm can carry out lossy to lossless coding of the image in the same architecture, which allows the lossless preservation of the ROI information. At last, two predefined quantization steps, one for the ROI and the other for the background, is used to quantize the transform coefficients. The quantization symbols are conditional entropy coded. The algorithm is of low computational complexity and achieves good compression results. The algorithm has the ability to code the ROI and the background with different predefined precisions. When the compression ratio is high, the restored ROI still has high fidelity.