【作者】 车丹丹；

【导师】 苗振江；

【作者基本信息】 北京交通大学 ， 信号与信息处理， 2010， 硕士

【摘要】 影视特效体现了视觉技术和艺术相结合的新发展,已成为当代电影的一种有效的表现手法,是创意实现不可或缺的技术保证。近年来,视频特效处理技术层出不穷,本文对经典的视频抠像-合成技术和卡通化技术进行研究。视频抠像-合成技术是视频特效制作中必不可少的关键技术,广泛应用于媒体制作中。目前,抠像技术的研究相对比较成熟,为了达到实时、高质量抠像的目的,我们采用图像分割与抠像相结合的方案——基于二值分割的贝叶斯抠像方法。该方法主要步骤包括：基于背景建模的背景减除、贝叶斯估计生成Trimap、Graphcut分割、贝叶斯抠像。准确的抠像结果是完美合成的保障,根据抠像中提取的前景颜色信息和像素点不透明度值,将提取出的前景融合到新背景中,这样就完成了整个抠像与合成。实验证明,我们的抠像-合成方法可以得到高质量的处理结果,而且不透明度对于模糊物体(如头发,羽毛等)的提取非常有效,我们还采用分割与抠像并行处理的机制来保证实时处理。影视艺术的卡通化,满足了现代人对视觉新鲜感的需求,使影视创作迈进了一个全新的艺术阶段。在形式多样的卡通动画制作中,将实际拍摄的视频,通过卡通化特效自动生成卡通视频的技术越来越受到人们的关注。本文我们采用一种基于特征的视频图像抽象化方法处理视频,该方法主要包括：重要性图计算、RGB与CIE-Lab颜色空间转换、快速双边滤波、亮度值伪量子化、各向异性的高斯差分线条抽取及线条增强。该方法的每一步都可以在图形卡硬件中并行处理,可以自动、快速的生成卡通视频。此外,我们将重要性图计算应用到卡通化处理中,保证了对重要性区域的细致刻画,符合人们的视觉习惯。我们的卡通化方法还有效地解决了现有视频和图像抽象化方法存在的抽取线条生硬,不连贯的问题,为非专业人员提供了一种简便直观的视频和图像的抽象化方法。本文中我们还应用以上两项技术的结合处理视频,即将视频中的前景(如人物)提取出来,然后对前景元素进行卡通风格化处理后,再合成到全新的场景中,这样也使得视频制作更加多样化。在未来的影视制作中也会得到越来越广泛的应用。本文的创新之处在于引入了重要性图计算到卡通化中,并将抠像-合成与卡通化处理相结合进行视频特效制作。更多还原

【Abstract】 Video special effect shows the combination of visual technology and art new development. It has become an effective expression technique of modern film and been an essential technology to ensure the creative implementation. In recent years, video effects processing technology emerge in endlessly. In our paper, we study two classic special effects technology-video matting-compositing and video cartoon.Video matting-compositing is a key technology in film production. In order to meet the requirements of real-time and high quality processing, we combine image segmentation and matting options-Bayesian video matting based on binary segmentation. The first step employs binary segmentation to generate a trimap. Given the trimaps, matting algorithms are then applied in the second step to refine the foreground boundary and extract alpha matte. The whole video matting process is automatically performed on PC.Experimental results on various test videos show the effectiveness of our method.Television cartoon art meets the needs of visual freshness, expands the creative way of film art. Our paper employs anisotropic difference-of-Gaussian (DoG) line extraction method based on feature-flow, which includes six main parts:importance map calculation, RGB and CIE-Lab color space conversion, fast bilateral filtering with bilateral grid, pseudo-quantization, anisotropic DoG, line enhancement technologies. Every step of the algorithm can conduct parallel processing on the PC graphics card. As our video extraction is based on anisotropic algorithm, it resolves the problem that the extracted lines are stiff and incoherent. In addition, we calculate importance map to ensure the detailed description for important area. This can generates a layering screen to meet people’s visual habits.In our paper, we apply the combination of above two technologies to process videos: extract foreground from real video, do cartoon-style processing, then synthesize foreground into a new background. This makes the film elements more diverse.更多还原