【作者】 王福龙；

【导师】 余英林；

【作者基本信息】 华南理工大学 ， 通信与电子系统， 2001， 博士

【摘要】 数字视频压缩编码技术是实现多媒体通信的关键技术之一。图象序列不仅表现在帧内象素之间的相关性，而且帧间象素之间也同样具有高度的相关性。本文对运动图像从三维的角度将图象序列视为立体图象，通过三维离散余弦变换(3DDCT)或三维小波变换(3DWT)变换后进行变换域的低码率压缩展开了较深入的研究。本文的主要内容包括： 1．对3DDCT算法进行深入研究，结合XYZ编码的特点，提出部分3DDCT算法及相对应的编解码流程，减少了运算量，提高了编码效率。 2．针对XYZ中量化策略进行了研究和实验，结合人眼视觉特性通过进化算法得出一个改进的量化表，该量化表可以有效提高编码效率。 3．针对XYZ编码中仍然使用2DDCT的RLE+Huffman的编码策略问题进行了研究，提出改进的RLE+Huffman编码策略，新的编码策略能较适应3DDCT的特点，实验表明，可以提高编码效率。 4．针对运动剧烈的视频序列XYZ编码效率下降的问题，提出结合运动补偿的3D MCDCT编码算法，将视频序列沿运动轨迹进行XYZ编码，使编码效率得到显著提高，但编码的复杂度加大，不易于实时应用。 5．数字视频序列可以看作是从模拟视频信号中抽样得到，帧间相关性过强是由于采样过密造成的；而帧问相关性太弱是由于采样频率太小造成的。本文从采样定理的角度提出变速XYZ视频编码，对帧间相关性太强的序列可以采用帧间跳帧的方法提高编码效率：而对帧间相关性太弱的视频序列可以采用帧间插帧的方法提高帧间相关性从而提高帧间编码效率。本文给出了定时长变速XYZ编码和变时长变速XYZ编码。 6．针对零树结构在多分辩图象编码中的有效作用，本文提出基于3DDCT的嵌入式三维零树结构视频编码，将3DDCT系数分解成多分辩率塔式结构，采用零树结构及算术编码，得到很好的编码效果。 7．对基于三维小波变换的视频编码进行了研究，结合三维零树结构及算术更多还原

【Abstract】 Digital video coding is one of the key technologies of multimedia communication The relationship not only exsits intrafiame , but also exsits interframe of video sequences . This thesis study low bitrate video coding through three dimension space: the video sequence regarded as three dimensional image is transformed by three dimensional diecrete cosine transform (3DDCT) or three dimensional wavelet transform(3DWT), then coded in the transformed space. The main work includes:1. The 3DDCT algorithm is deeply studied in this thesis . Combined the characteristic of XYZ coding, a video coding program based on part 3DDCT algorithm is proposed, which can raise code and decode efficiency by reducing the caculation of transform.2. The quantize policy of XYZ video coding is researched . Combined the visual characteristic of human eye, a new quantization matrix on XYZ video coding is given through evolutionary algorithms. The simulation results show that this method can give high compression ratio and good reconstruct picture qanlity.3. The code policy RLE+Huffman using in the XYZ video coding is still the policy based on 2DDCT, which is not suited to XYZ coding based on 3DDCT. A new code policy is proposed in this thesis which is more suitable for XYZ coding. Experiment results show this new method has higher performance than previous ones.4. The XYZ compression efficiency decrease in case of video sequence with violent motion. To counter video sequences with violent motion, a new video coding scheme based on three dimensional cosine transformation with motion compensation is presented. Every eight oringal sequence pictures are compensated through motion estimation, and then use XYZ method to code. The simulations demonstrate that this method is very efficient.5. Digital video sequence can be regarded as sampling of analogue video signal, sampling too intensive can make the intraframe relationship higher, and sampling too sparse can make the intraframe relationship lower. Based on sampling theory, a variant rate XYZ video coding method is proposed in this thesis, which can improve the coding efficiency.6. Zerotree coding is very efficient in multiresolution image coding. With reorganize the更多还原