【作者】 庄怀宇；

【导师】 吴成柯；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2006， 博士

【摘要】 基于小波变换的内嵌编码技术已成为当前静止图像编码领域的主流，目前最优秀的内嵌编码算法是优化截取的内嵌码块编码（Embedded Block Coding with Optimized Truncation，EBCOT）算法，其由Tier-1和Tier-2（简称T₁、T₂）编码器构成。算法具有码流可随机获取以及良好的恢复图像质量等众多优良特性，因此成为实际应用中首选算法。 在静止图像编码领域里，高性能以及低复杂度编码算法是人们追求的目标之一。本文针对优化截取的内嵌码块编码算法中码率控制部分作了深入研究，并致力于高性能、高效率的码率控制算法及其实现结构的研究，提出了具有创新性的降低计算量、存储量，易于硬件实现的高性能码率控制算法，并成功应用于图像编码硬件系统中。本文主要的工作成果可以概括如下： 1．提出了一种易于硬件实现的T₂编码器快速算法并详细给出了其硬件结构。通过降低率失真斜率估计的计算复杂度、简化优化截取方法和码流组织中Tag Tree编码方法，降低了T₂编码器硬件实现的难度，减少了硬件资源，提高了JPEG2000硬件系统处理的并行度，输出码流符合JPEG2000标准格式，图像质量下降很小。 2．提出了一种基于重要性权值和反馈控制编码深度的率失真优化算法。对小波变换后不同比特平面的编码过程（Coding Pass）赋予不同的重要性权值，根据重要性由高到低依次进行熵编码，同时T₂编码器根据已得到的比特率自适应地反馈控制T₁的编码深度。在保持图像质量的前提下，本方法有效地减少了EBCOT算法的计算量和内存使用量。 3．提出了一种新的干涉多光谱图像ROI（感兴趣区域）编码小波分解结构。根据光谱特性，对高频系数光谱ROI区域进行垂直方向的分解，背景区域不进行分解，使光谱信息的能量更加集中，提高光谱区域的编码效果。 4．提出了一种新的基于码块预测的JPEG2000多图像片压缩的码率预分配算法。对各个图像片小波变换和量化后EBCOT编码码块的有效比特平面进行独立的熵估计，并以图像片估计熵总和指导码率预分配。算法复杂度低，熵估计精确，重建效果十分接近全局压缩效果。 5．设计实现了适合于卫星图像传输符合SPIHT算法的小波变换及空间生成树硬件编码器和符合JPEG2000标准的T₂硬件编码器。这两套编码器有较高系统处理速度，并且提供通用接口，可以很方便地与其他系统连接，可应用于不同场合。更多还原

【Abstract】 Embedded wavelet coding technique is a main method for image coding, in which, EBCOT （Embedded Block Coding with Optimized Truncation） is an algorithm with highest performance at present, which is composed of Tier-1 and Tier-2 （short for T₁ and T₂） encoder. Because of good restored image and random access for coding stream, the algorithm becomes the first choices in practice.In the image coding field, the design of high performance image coder is always the goal that many relative researchers pursue. In the paper, Rate control of embedded block coding is studied in deep, especially on rate control algorithm and its implementation with high efficiency and performance. The creatively rate control algorithmes with low computation, low memory usage and high efficiency for hardware implementation are presented in the paper, which have been applied to real coding system successfully. The followings are main content of this paper:1. A fast algorithm for T₂ encoder in JPEG2000 suitable for hardware implementation is presented in this paper. By reducing the calculational complexity of the rate-distortion slope estimate and simplifying the design of optimized truncation and Tag Tree encoding in code stream organization, the difficulty and the resource of hardware implementation for T₂ encoder is reduced, and the parallelizability of the JPEG2000 hardware system is enhanced. The final code stream is accorded with the standard format of JPEG2000 and the loss of image quality is little.2. An adaptive rate control algorithm based on significant priority is presented for embedded block coding with optimized truncation. The coding passes of all coding block are evaluated by different significant priority according to the bit-planes the pass located in and the coding type, which are encoded by T₁ encoder in the significant priority order. T₂ encoder sends feedback to control the coding depth by the accumulated bit rate and complete the rate-distortion optimization truncation. The proposed scheme can efficiently reduce the cost of computation and memory usagc, which is suitable for hardware implementation.3. A novel wavelet decomposition framework is presented for multi-spectral image coding based on ROI Coding of EBCOT. After normal wavelet transform, the vertical wavelet decomposition is introduced for ROI which include the multi-spectral information in the high frequency subband, not for the background region. It can concentrate the enegy of spectra and enhance the coding result of ROI.4. After the elaborate analysis for the problem of rate pre-allocation for the remote sensing image JPEG2000 multi-image tiles encoding, a novel rate pre-allocation algorithm based on code block prediction is presented to eliminate influence of the reconstructed image quality using equal allocation. After wavelet transform and quantification, a prediction module is introduced to estimate the entropy of EBCOT code block within available bit plane, then using the summation estimate entropy of each image tile to allocate the rate for each image tile. The proposed algorithm matches the code mode of JPEG2000, with little loss of images quality especially for the image whose object has the characteristic of precinct convergence.5. Wavelet transform hardware module, space orientable tree hardware module and T₂ encoder hardware module for SHPIT and JPEG2000 are implemented. And a common net interface is implemented in coder for data exchange with external network.更多还原