【作者】 赖兆强；

【导师】 田金文； 谭毅华；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2006， 硕士

【摘要】 成像光谱技术是上世纪80年代发展起来的新技术,集中了光学、光电子学、电子学、信息处理、计算机科学等领域的先进技术,它将成像与光谱技术结合起来,能够在连续光谱段上对同一地物同时成像。随着成像光谱仪的普及应用,图像的空间分辨率、谱间分辨率、时间分辨率越来越高,使得成像光谱数据量迅速增长,高光谱图像压缩技术是数据存储和传输中的一个迫切需要解决的问题。从信息论的角度来看,所有的压缩技术都是通过去除冗余来达到压缩目的。高光谱图像存在两类冗余:空间冗余和谱间冗余.其空间冗余表现为,在属于同一类地物的相邻象素之间存在空间相关性;其谱间冗余表现为,由于谱间分辨率的提高,相邻波段同一位置的象素之间存在相关性。本文首先简要地介绍了高光谱技术以及图像压缩编码技术。接着从信息论的角度介绍了JPEG2000的技术基础,对小波变换,算术编码等进行了描述。然后详细分析了基于JPEG2000框架的编解码系统,对小波提升算法和位平面编码等各个模块展开讨论。接着阐述高光谱图像压缩方法,介绍了两种分组方法:分段的DPCM和C-MEAN聚类;分组后,结合基于JPEG2000框架的编解码系统,我们实现了高光谱图像压缩算法。采用DPCM预测方法去除谱间冗余,预测得到的差值图像首先进行小波变换,再用位平面编码压缩,去除空间冗余,最后形成码流输出。实验结果表明,基于JPEG2000框架的高光谱图像压缩算法,对高光谱图像在8倍压缩比的情况下,能取得较好的效果。更多还原

【Abstract】 Imaging spectral technique is a new technology developed in 1980s. This technology is able to imagine the same object in the continuous spectral band by integrating the imaging technology and spectral technology, as well as optics, optoelectronics, electronics, information-processing and computer science technology. With wide application of the imaging spectrometer, images’ spatial spectral and time-resolution become higher, which correspondingly improves the spectrum data largely. The compression technology of hyperspectral image is a key problem to be solved in the storage and transportation of data.From the aspect of entropy theory, all compression technologies are to reduce redundancy. Two kinds of redundancy exist in hyperspectral image: space-redundancy and the redundancy between spectrums. Space-redundancy can be found among the pixels in the same kind of objects; Redundancy between spectrums can be found at the same place of spectrums due to the increase of the resolution.This thesis introduces hyperspectral technique and image compress coding technique. It makes a introduction of JPEG2000 and describes the wavelet transform, arithmetic coding and so on. And then we amply analyze the coding system which bases on the JPEG2000 and the modules of lifting wavelet transform and bit-plane coding. During the section of hyperspectral image coding system, we adopt two methods: regrouped DPCM and CMean; after grouping the images, combining the coding system, we implement the hyperspectral image coding system. The method based on regrouped DPCM includes two aspects: regrouped DPCM for in-frame decorrelation and image-compression based on wavelet transformation with bit-plane coding exporting the coding stream. The fine experimental results demonstrate that the proposed algorithm is effective and practical with the goal of 1/8 compression.更多还原