【作者】 张培强；

【导师】 沈兰荪； 张晓玲；

【作者基本信息】 北京工业大学 ， 电路与系统， 2004， 硕士

【摘要】 摘 要 高光谱图像是由成像光谱仪在不同光谱波段获得的序列图像，在二维遥感图像的基础上又增加了光谱维的信息。这种高分辨率、连续谱段的图像具有极大的数据量，对传输和存储带来了不便。为此，针对具体应用的数据压缩技术被广泛采用，包括有损压缩、近无损压缩和无损压缩技术。 由于高光谱图像主要用于对地物进行分类、识别、监测等等，压缩过程应尽量不丢失有用的信息。论文主要研究高光谱图像的近无损和无损压缩技术，侧重于无损压缩算法。另外，为了便于图像的检索和浏览，应能够提供从有损到无损的具有渐进传输性能的压缩码流。 算法的研究以小波变换为基础，利用提升方法构造整数可逆小波变换，并根据高光谱图像自身的特点，选择适合的整型小波变换，结合等级树集合分裂(SPIHT)算法和集合分裂嵌入式块编码(SPECK)算法，利用位平面编码的思想实现波段图像的渐进传输，最终得到无损压缩的结果。此外，根据光谱波段之间的相关性分析结果对若干波段进行分组，将各组波段图像组织成数据立方体的形式进行三维小波变换，在去除图像空间相关性的同时，也去除了波段图像之间的相关性，从而进一步降低图像的熵值。再运用 3-D SPIHT 算法对三维小波系数进行编码，实现了无损压缩，相比直接使用二维方法压缩，压缩比有了进一步提高。由于光谱维相关性很强，进行波段分组后，先在光谱方向进行小波变换，去除波段图像之间的相关性，然后再单独对各个波段使用 SPIHT 或 SPECK 算法，也获得了很好的无损压缩结果。 论文还初步探索了基于小波变换的视频编码算法用于光谱图像的压缩，取得了初步的结果。更多还原

【Abstract】 Hyperspectral image, an image sequence generated by spectrometer in manynarrow ranges of wavelength, possesses additional spectrum informationcorresponding to traditional two dimensional remote sensing images. The character ofthe high spatial resolution and high spectral resolution makes the hyperspectral imageoccupy huge set of data, which results in heavy burden for data transmission andstorage. Therefore, some application-specific data compression techniques should beapplied, including lossy, near-lossless and lossless compression. Hyperspectral imaging is a promising technique and mainly used for surfacedetection and identification, target classification and status monitoring. Thecompression scheme should not lose useful information in original data. In this paper,we study related lossless and near-lossless compression for hyperspectral image.Moreover, the compression scheme provides embedded bitstream from lossy tolossless, which facilitates the image retrieving and browsing. The proposed algorithms here are based on the wavelet transform. For losslesscompression, the lifting scheme constructs the reversible integer wavelet transform.The choice of the wavelet type based on the analysis of the character of hyperspectralimage, then, SPIHT or SPECK algorithm will be applied for bitplane coding that canrealize the lossy to lossless progressive transmission. We also exploit the correlationbetween adjacent bands, the result implies that group of bands can form rectangularprism, thus the three dimensional transform can act on it, causing decreasedcorrelation both in spatial dimension and spectral dimension, finally 3-D SPIHTalgorithm sorts the wavelet coefficient along the path of 3-D trees. The result oflossless algorithm has slight improvement than that of 2-D algorithm. The mostsurprising way is not the 3-D algorithm, but the 2-D algorithm used for decorrelatedgroups of bands, i.e. applying wavelet transform first on spectral direction, then 2-Dalgorithm is used to explore the coefficients’ relation. The experiment gives thestirring result for lossless compression. Video coding methods based on wavelet transform are introduced in this paper forhyperspectral image compressing, the result is acceptable for near losslesscompression but still have some other disadvantages such as highly computingcomplexity. The study here is elementary and still has a long way to go.更多还原