节点文献

基于小波理论的SAR图像压缩算法研究

Research on Wavelet-Based Sar Image Compression Algorithm

【作者】 王爱丽

【导师】 张晔;

【作者基本信息】 哈尔滨工业大学 , 信息与通信工程, 2008, 博士

【摘要】 合成孔径雷达(SAR)是一种主动式的微波成像系统,可安装在飞机、卫星等飞行平台上,全天时、全天候地实施对地观测,并具有一定的地表穿透能力,在军事和民用建设中发挥了巨大作用。通常SAR图像分辨率较高,成像范围较大,数据量高,因此如何有效地对其进行压缩,降低数据量,以利于存储和传输便成为SAR图像处理中的一个关键问题,具有重要的理论意义和应用价值。围绕着SAR图像的特点,本论文重点研究了小波域的SAR图像编码算法,主要包括以下内容:首先,论文对SAR图像特性做了较深入的理论分析。第一是受乘性相干斑噪声的影响,SAR图像数据相关性小,信息熵高;第二是SAR图像上既有细节纹理信息又有大量均匀区域,这样就有必要考虑减少均匀区域的编码比特数;第三是SAR图像数据的动态范围很高。SAR图像与光学图像之间的这些差异意味着将光学图像的编码算法用于SAR图像压缩并不一定是最优的,需要结合SAR图像特点设计相应的编码算法。接着,针对SAR图像空间相关性低,论文研究了基于小波变换的SAR图像编码算法。基于后续SAR图像应用前多进行相干斑去噪的预处理,论文在多级树集合分裂(SPIHT)编码算法基础之上,结合空间树结构(SOT)进行相干斑去噪后再编码,改善了重建图像质量;提出了结合小波树构建矢量进行量化、编码的SAR图像压缩,去除了空间矢量量化的块效应,实现了大压缩比下SAR图像压缩;最后,针对SAR图像的目视判读,在小波域引入人类视觉系统(HVS),对人眼认为重要的信息进行加权,去除了编码中的视觉冗余,改善了重建SAR图像的主观视觉质量。然后,论文研究了基于多小波变换的SAR图像压缩。大多数单小波无法同时兼顾正交性和对称性,而多小波能同时满足对称性、紧支性、消失矩和正交性的要求,在信号处理方面比单小波更有优势。针对多小波分解系数的特点,提出了基于多小波的改进SPIHT算法用于SAR图像压缩,获得了优于传统SPIHT编码算法的重建图像质量。同时,探讨了在多小波域编码前进行改进的软阈值相干斑去噪,抑制相干斑噪声的同时尽量保持图像边缘信息,实现了多小波域去噪和编码相结合,改善了图像重建质量,两幅测试图像的重建峰值信噪比分别提高了1.5dB和1.0dB。最后,针对SAR图像含有丰富的纹理信息,论文研究了基于小波包分解的SAR图像压缩。SAR图像的纹理信息多分布于中、高频子带。小波变换只对图像的低频子带进行分解,而小波包变换对图像的高频子带也进行分解,能够更好的与信号能量分布情况进行匹配。因此,论文研究了小波包分解最优基选取的代价函数设计问题,提出与后续编码算法结合的量化比特数作为代价函数,并根据子带能量分布特性进行重要性加权,实现了非均匀量化,更好地保留了SAR图像的纹理信息。

【Abstract】 Synthetic aperture radar (SAR) is an active imaging microwave sensor which can be carried on a variety of airborne and spaceborne platforms. SAR is envolving to become an indispensable reconnaissance tool for military purposes because SAR can work under any times and any weathers. In the last few years, high spatial resolution images of the earth produced by SAR systems have large imaging size and the collection capacity for SAR images is increasing rapidly, thus SAR image transferring and restore become very challenging problems. Therefore, the research of SAR image compression with high fidelity is of great theoretical significance and application value, and is also an urgent task in the research and development of the SAR system.This thesis mainly studies the SAR image compression algorithms in wavelet domain focusing on the unique features of SAR image. Main work includes four aspects below:Firstly, the characteristics of SAR image are analyzed in detail, which affect the design of image compression algorithm. The first is speckle noise which reduces space correlation of image pixels, increases information entropy, and severely depresses SAR images quality. The second is that SAR images have both detailed texture information and many uniform regions. It is necessary to reduce bit rate of uniform regions. The last is that SAR image has high dynamic range This kind of remarkable difference means that those encoding/decoding algorithms for optical image data is not optimal for SAR data. It is necessary to design coding methods for SAR image combing with its unique characteristics.Secondly, SAR image compression algorithms based on wavelet transform are proposed. In following SAR image applications, speckle noise is first reduced. Based on the set partitioning in hierarchical trees (SPIHT) algorithm, speckle noise removal using spatial orientation trees (SOT) is introduced before coding to improve the quality of reconstructed SAR image; According to SAR image compression at high compression ratios, vector quantization (VQ) of wavelet trees decreases the block effects of VQ in space domain; At last, we study the effect of incorporating a human visual system (HVS)-based transform model in SPIHT algorithm to reduce visual redundancy and improve subjective perception quality for visual interpretation appliations.Thirdly, SAR image compression based on multiwavelet transform is proposed. Multiwavelet can possess desirable features simultaneously, such as the finite support, symmetry and orthogonality, while wavelet cannot. Thus multiwavelet has more advantages than wavelet in signal processing. Accroding to the feature of multiwavelet coefficients, we propose modified multiwavelet-based SPIHT algorithm to compress SAR image and obtain better reconstructed image quality than wavelet-based SPIHT algorithm. At the same time, we introduce modified soft-thresholding denoising method which suppresses the speckle noise while keeping edge well before image coding. Thus denosing and coding are joined in multiwavelet domain for SAR image to improve the reconstructed image quality. The peak signal to noise ratios of the reconstructed images are improved 1.5dB and 1.0dB.Last, SAR image compression methods based on wavelet packet are studied. SAR image has rich texture information which distuibute in middle and high frequency subbands. Wavelet transform just decomposes the low frequency subband, while wavelet packer transform also decomposes high frequency subbands and matches the energy distribution of signal better. We study the cost function of best base and propose the actual quantization bits as cost function combing the following coding method. Then different subbands are weighted according to importance and nonuniform quantization is realized to keep texture information of SAR image better.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络