【作者】 张佳岩；

【导师】 周廷显；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2007， 博士

【摘要】 基于小波变换的图像压缩算法具有分辨率及保真度可分级性(从有损到无损渐进传输)、较强抗误码性能、低内存空间占用和编解码快速等特点。它能有效缓解图像在深空通信(DSC)存储和传输中,大量占用整个系统有限缓存及带宽资源的矛盾,提高深空探测器处理图像信息量的能力。但DSC传输数据与通常的地面和卫星通信相比有更多的挑战,诸如长距离、非常低的信噪比、高信号衰减延迟和数据毁坏率、不对称带宽等。因此不能简单地把Internet网上的基于小波变换的图像压缩算法(如JPEG2000)直接用于DSC中。为此,本文以CCSDS在2005年推出了应用于DSC的基于小波变换的图像压缩标准(简称CCSDS标准)为基础,以DSC与Internet网传输压缩图像中的编解码需求、容错性和数据用途3个不同点展开研究。开始详细介绍了该标准,并通过与JPEG2000算法比较发现,二者虽然有同样的基本结构,但它们之间的小波基、小波变换级数、位平面编码及不同小波系数之间相关性处理都有很大不同。这些不同主要为了降低算法复杂度,从而能减少探测器的功率消耗和寄存器的使用。通过对DSC压缩数据流存储及传输的特殊性问题研究,本文提出适合CCSDS标准特点的4种方法。首先提出一种适合CCSDS标准的对整幅图像分块的方法。它能进一步降低探测器缓存溢出及压缩数据流传输过程中的干扰造成地面解码器的错误传播。分割在图像小波变换域进行,在预分割的基础上,各个分块再进行基于图像特征的分割及融合。各个分块是基于多种图像特征(如颜色,频带及纹理等)通过高斯混合模型完成分割,并可根据探测器CPU使用情况选择是否采用ICM及朴素Bayes法则实现联合特征分割。用仿真对分割的结果验证。其次提出一种CCSDS标准感兴趣区压缩的算法,并根据标准的速率-失真(R-D)模型及借鉴一种用于语音量化最优分配位的方法,实现了感兴趣区(ROI)与非感兴趣区(NON-ROI)的位优化分配。它根据图像的优先权分区及压缩算法中位平面编码特点,调整ROI与NON-ROI小波系数间位平面相对位置。且为图像分割的每个区域提供最优压缩比,使固定的有限速率下行信道能优化分配发送比特。其CCSDS标准的R-D模型则是基于Mallat关于图像变换域R-D关系的启发,对多幅图像的测试仿真所得。然后提出一种对CCSDS标准树形的小波系数重新分组的压缩方法,并根据标准的R-D模型及借鉴一种用于语音量化最优分配位的方法,实现不同分组位优化分配。这种方法不但能实现压缩数据流的鲁棒性传输,且也能达到地面接收图像信息更大化的目的。它把小波系数分成9组独立、交织的位流,并按照最终的信息量最大化原则为每组分配一个优化压缩比。最后提出一种CCSDS标准与Turbo码联合解码的方法。它对压缩端没做任何改变,但在信道信噪比不变的情况下,却能降低接收压缩数据流的错误率,该方法尤其适合DSC。它根据CCSDS标准熵编码过程中的残留冗余、不同小波系数之间的相关性,结合Turbo码解码过程中迭代方法、分量码MAP解码中对外信息的使用,实现最终的信源信道联合解码。更多还原

【Abstract】 The image compression algorithms based on Discrete Wavelet Transformation (DWT) can provide ratings according to the image resolution and fidelity (the progressive transmission from lossy to lossless).And it has stronge anti-mistake coder ability, low memory space taken up and fast coding characteristics. Therefore it can availably abate the conflict that the storage and transmission of images in Deep Space Communication (DSC) largely occupy the limited buffer and channel bandwidth of the whole system, and improve the ability that the explorer deals with the images.But compared with common terra and satellite communications, DSC faces more challenges, such as long distance, very low signal noise ratio, high signal propagation delays and data corruption rates, asymmetric bandwidth and so on.Thus, the image compression algorithms based on DWT in the Internet (such as JPEG2000) can not simply apply to DSC.As for it, this paper takes the image compression standards based on DWT that the CCSDS(referred as CCSDS standards) released in 2005 as foundation, studies three dissimilarities of transmitting the images between DSC and the Internet according to user requirements,error containment and data purpose.In the beginning, the CCSDS standards based on DWT is studied in detail.And compared with JPEG2000, although their basic structures are the same, there are different highly about Wavelet Base, DWT Level, Bit Plane Coding and the relativity between the coefficients. As these dissimilarities can lower complications, the power and the usage of the buffer can be lowered on the explorer.Based on studying the special problem that DSC transmit and store compressed bit streams, this paper presentes four methods fit for the characteristic of CCSDS standards.Firstly, the image segmentation method is presented that is fit for the CCSDS standards.It can lower the influence that buffer overload and error propagation of the ground decoder because of receiving error compressed bit streams. The segmentation is carried on after image DWT. After pre-segmentation of the image, each block is divided up and fused again.Each block can keep on be partitioned based on the image features (such as color, band and texture etc.) by Gaussian mixture models. Based on the usage of CPU in the explorer, ICM and naive Bayes can be selected to finish the segmentation together.The results are verified by simulation.Secondly, the new image compression scheme based on Region of Interest (ROI) is presented that is fit for the CCSDS standards.And according to the Rate- Distortion(R-D) model of CCSDS standards and the method used for voice quantization optimum bit assignment, the optimum bit assignment between ROI and NON Region of Interest (NON-RON) is achieved.Based on the priority partition table of the image and the character of Bit Plane Coding, the opposite position between ROI and NON-ROI is adjusted. Compression rate of each block can be assigned in order to carry out the optimized assignment of transmission bits under the condition of limited and fixed download channel rate. According to the elicitation of the relationship about R-D given by Mallat, R-D model of CCSDS standards is obtained by simulation of several test images.Thirdly, the method fit for CCSDS standards is presented.It divides wavelet coefficients of tree structure to different groups. Based on R-D model of CCSDS standards and the method used for voice quantization optimum bit assignment, the optimum bit assignments between different groupings is achieved.It can not only achieve robust transmission of compressed bit streams, but also make the information of image received by the ground more. It groups the wavelet efficients and produces nine independent and interlaced bit streams, and assigns compression ratio for each group so that total information can be maximized.Finally, the method of the joint decoding between CCSDS standards and Turbo coder is presented. When signal to noise ratio is invariable, the bit error rate of receiving the compressed bit streams can be lowered even if compression coding do not change at all. Such method is especially fit for the deep space communication.By combining the residual redundancy during entropy encoding and the relativity between the different wavelet coefficients of CCSDS standards with the iterative and the usage of extrinsic information during Turbo decoding, joint decoding can be carried out at last.更多还原