【作者】 陆悌亮；

【导师】 龚声蓉；

【作者基本信息】 苏州大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 随着多媒体技术的不断成熟,人们对多媒体信息的处理方式产生了新的、更高的要求,更加注重多媒体系统的交互性和灵活性。传统的视频编解码技术是基于帧的,而MPEG-4编解码标准所采用的方法是基于对象的,就是对视频划分为不同的、连续运动的视频对象,对每个对象单独进行编码。每个对象都包括三种信息:形状信息、纹理信息和运动信息。对视频对象的编解码就需要对这三种信息进行编解码。其中MPEG-4中的核心技术之一就是形状编码部分,正因为采用了形状编码技术,MPEG-4才能够提供与众不同的能力——基于内容的交互性。目前,国内外对MPEG-4中的视频对象分割和形状编码研究都是相互独立的,没有考虑两者之间的关联性。因此,本文在国内外现有形状编码研究成果基础上,对视频对象分割和形状编码彼此的结合做了深入研究,主要工作包括:1、研究了基于形变模型的图象分割技术,对参数形变模型和几何形变模型技术做了分析比较,并深入研究了一种几何形变模型技术——水平集方法。2、提出了基于水平集的运动视频对象分割方法。根据运动视频系列的特点,通过相邻帧的亮度差得到初始轮廓,以此轮廓为初始零水平集,采用窄带水平集方法,分割出运动对象。3、提出了新的基于基准线的形状编码算法,针对现有基准算法中距离集和拐点采样算法的不足,提出了能更好的适应各种边界走向的新算法。4、提出了结合视频对象分割的形状编码新体系,现有MPEG-4中,视频分割到形状编码的整个中间过程造成了一定的时间花费与误差,提出了在视频分割后,直接进行形状编码的新思想,可以有效的避免从分割到编码的中间过程造成的时间花费与误差。更多还原

【Abstract】 Along with the unceasing maturity of multimedia technique, people have advanced new and higher requirements for the processing mode of multimedia and paid more attention to the interactivity and flexibility. The conventional video coding/decoding technique adopts frame-based coding technique, while MPEG-4 coding/decoding standard adopts object-based coding technique. MPEG-4 standard divides video into different、consecutively moving video objects and codes every object separately. Each object includes three kinds of information: the shape information、the texture information and movement information. The three kinds of information of every video object are needed to code. One of the core techniques in MPEG-4 is the part of shape coding. By reason of having adopted the shape coding technique, MPEG-4 has the larruping ability—the content-based interactivity.The research on the video object segmentation and the shape coding is always isolated. Based on the progress of the research on the shape coding, the combination of the video segmentation and the shape coding is researched carefully. The main work includes:1、The image segmentation technology based on Active contour models is researched. The parameter models and geometrical models is analysed and compared, and a geometrical models, level sets methods, is researched especially.2、A moving object segmentation algorithm based on the level set methods is proposed. Take the luminance difference as the primary zero level set, and segment the object by narrow band algorithm.3、A new shape coding based on the base-line is proposed. Since the shortness of the current algorithm, proposes a new algorithm which can adapt to any direction of the outline, as well as a new decode method.4、A new system which is called shape coding system combined with the object segmentation is proposed. In MPEG-4, the middle process results in too more time cost and errors, so proposes a new system, which coding shape jusr after the object segmentation. The new system can save the time and the errors.更多还原