【作者】 马冬梅；

【导师】 马彩文；

【作者基本信息】 中国科学院研究生院（西安光学精密机械研究所） ， 光学工程， 2009， 博士

【摘要】 高光谱图像在采集地物空间信息的同时,亦获取地元的光谱信息,从而提供更为丰富的地物细节,因此在农业、地质、海洋、军事等领域有非常重要的应用价值。随着成像光谱仪分辨率的不断提高,高光谱图像数据量剧增,给数据的存储和传输带来了很大困难,因此必须对其进行压缩。高光谱数据作为一种三维图像,不同于二维静止图像,也不同于视频图像,一般的图像压缩技术难以达到高光谱压缩的性能要求。因此分析高光谱图像本身的特点,研究适合高光谱图像压缩的有效算法,具有重要意义。论文详细分析了大孔径静态干涉成像光谱仪(Large Aperture StaticImaging Spectrometer,LASIS)高光谱干涉图像的特点,并得出了相关的结论。LASIS图像不仅具有普通静止图像的帧内相关性,同时存在较强的帧间相关性及谱间相关性。LASIS高光谱干涉图像具有谱像合一的特点,0光程差谱线附近光谱条纹明显,富含光谱信息,需要充分保护,这是LASIS图像的最明显的特性,也是压缩技术的难点之一。后续压缩方法的研究主要围绕LASIS图像的特性展开,以期获得较好的压缩效果。论文针对LASIS图像压缩的高实时性、低复杂度、低存储资源等要求,提出了改进的无链表SPIHT算法(MLSPIHT)。该算法采用2bit的系数状态标志代替SPIHT算法的3个可变长度链表,省却了存储链表需要的大容量不定额外内存及复杂的链表操作。相对于无链表零树编码LZC算法,进一步降低了存储需求和算法复杂度。根据LASIS图像序列本身的特点,提出以下几种压缩方案:(1)、3D非对称等长树DWT和基于感兴趣谱段BOI的3D-MLSPIHT算法。该算法以LASIS系统推扫过一个完整视场形成的图像序列为编码单位,充分利用LASIS图像的谱间相关性,在给定的压缩比下,取得了最佳压缩性能。(2)、基于感兴趣区域ROI的3D-MLSPIHT算法,该算法以连续8帧图像为编码单位,适合于高实时性、低内存的应用环境。(3)、基于感兴趣区域ROI的准三维MLSPIHT算法。该算法以相邻两帧图像为编码单位,进一步降低了内存需求和计算复杂度,其压缩性能介于3D-SPIHT和2D-SPIHT算法之间,适合于对实时性和内存容量要求苛刻的应用条件。以上算法在8:1的压缩比下,PSNR值大于40dB,同时有效的保护了光谱信息,并满足LASIS高光谱图像压缩系统的应用要求。本论文从分析LASIS高光谱干涉图像的特性出发,围绕LASIS图像所特有的光谱特性,针对不同的应用环境,提出了相应的压缩算法,并对所提出的方法进行了实验分析和性能评价。更多还原

【Abstract】 Hyperspectral imagery is generated by collecting spatial information as well as the spectral information of the earth targets, and describes the targets with more detail, so it has great value of applications in agriculture, geology, ocean and military surveillance. With the increasing of remote sensor resolution, we get massively large image data sets which it is difficult to access and transport. Therefore they must be compressed before being processed, storage and transmission. As a kind of three-dimensional data sets, it is different from the 2D still image, and is also different from video series, so the general image compression method is not efficient for them. Analyzing their character and study appropriate algorithm for coding is very necessary.We particularly analyze the character of the Large Aperture Static Imaging Spectrometer (LASIS) hyperspectral interference images and give the inclusions. LASIS images not only have the spatial correlation, which is the same as general 2D still images, but also have the inter-frame correlation and inter-band correlation. LASIS image is amplitude modulated by the interference spectrum. The spectral information concentrates about the zero optical path difference, and needs to be efficiently protected. This character is very important, which makes it difficult to process. The followed compression algorithms specially designed according to LASIS’s character have good performance.In order to meet the demands of high real-time, low complexity and low memory, we proposed a Modified Listless SPIHT algorithm, which uses 2-bit state-flag per coefficient, instead of the three variable-length lists of SPIHT. The MLSPIHT save much memory and complicated list-operating required by the original SPIHT, and need less memory and operation than the LZC.Considering the characteristic quality of LASIS, we proposed the following scheme: (1) 3D asymmetrical DWT with equal length tree and 3D-MLSPIHT based on Band Of Interest. The scheme defines the whole LASIS sequences as a coding unit and take fully consideration of LASIS’s inter-band correlation. It get good performance under the given compression rate. (2) 3D-MLSPIHT algorithm based on Region Of Interest. This method define every 8 sequential images as a coding unit, and satisfies the requirement of less memory and lower operation complexity. (3) Nearly 3D-MLSPIHT algorithm based on Region Of Interest. This algorithm defines tow contiguous images as a coding units to decrease more memory and operation complexity. Its compression performance is between the 2D-SPIHT and 3D-SPIHT. This scheme is specially designed under the condition of the highest real-time and the lowest memory. All the algorithms have the PSNR of above 40dB at the 8:1 rate and efficiently protect the interference spectral information as well. They all meet the application requirement of the LASIS system.This paper is based on the analysis of the LASIS image characters, especially on its spectral character. According to different application requirements, we propose different schemes, and gives the numerical experimental results and conclusions.更多还原