视频风格化论文：视频风格化 SIFT Flow SBR算法 遮挡 非刚性运动.docVIP

【关键词】视频风格化 SIFT Flow SBR算法 遮挡 非刚性运动 【英文关键词】video animation SIFT Flow SBR algorithm shielded non-rigid motion 视频风格化论文：基于稠密特征对应关系的视频风格化关键技术与系统 【中文摘要】视频风格化技术是一种基于某一风格化效果，对视频的连续帧场景进行渲染的方法。我们对视频各帧直接进行图像风格化，渲染后的视频帧会出现画刷抖动，影响视觉观赏效果。视频风格化算法结合了视频跟踪技术，借助连续帧之间的变化关系来传播画刷，能很好地解决画刷抖动问题。本文提出了一种基于稠密特征对应关系的视频风格化算法。首先，我们采用改进的SIFT Flow算法对原视频中物体进行跟踪、对应，得出位移关系矩阵信息。接着，我们采用一种基于画笔渲染(Stroke-Based Rendering, SBR)的算法对视频第一帧进行图像渲染，生成画刷。最后，我们根据之前计算出来各视频帧之间的对应关系来传播画刷，实现视频的风格化。各种不同运动类型场景的实验结果表明，本文方法在处理受遮挡目标和非刚性运动目标的场景下有较高的准确率和较好的渲染效果。本文的改进SIFT Flow算法适用于任意的SBR算法，可实现同一段视频的多种风格化渲染。本文实现了一个基于稠密特征对应关系的视频风格化系统。系统包括5个功能界面，为用户提供交互分层、执行SIFT Flow计算等7种功能。在执行SIFT Flow计算后，用户可根据自己喜爱的画刷进行参数调节，对视频实现多种风格化的渲染。 【英文摘要】Video animation is a method for rendering continuous video frames based on acertain image rendering style. Directly rendering each video frame by one imagepainterly animation algorithm will cause strokes flickering. As a result, the strokes抐lickering affects our visual ornamental effect. Video animation algorithm combineswith the video tracking technology, spreads the strokes with the relationship viacontinuous frames, can solve the flickering problem well.This paper proposes a video animation algorithm based on dense featurecorrespondence. Firstly, we use improved SIFT Flow algorithm to track andcorrespond objects in original video and we can obtain the information ofdisplacement relationship matrix. Secondly, we apply one SBR algorithm to renderthe first frame of video, and we can create the information of strokes. Lastly, wespread the strokes with the displacement relationship matrix to achieve videoanimation. Experiments on a variety of videos show that this algorithm has a higheraccuracy and a better rendering effect in dealing with the shielded objects andobjects with a non-rigid motion. This improved SIFT Flow algorithm applies any SBRalgorithm. We can render one video by various video animation styles.This paper implements a video animation system based on dense featureco