一种基于分布特点的口播帧识别算法

针对现有的大部分口播帧识别方法均缺乏通用性,提出一种基于口播帧分布特点、无须模板的检测算法.首先,视频流被切分为镜头集合并提取满足一定条件的候选镜头;其次,对相似镜头进行聚类;最后,研究每个类的分布特征.根据口播帧时间分布特征,检测出包含口播帧的聚类集合.实验表明,本文算法实用性强,应用范围广,准确率基本达到100%.     

新闻视频单元高效切分方法的研究与实现

提出了一个基于口播检测的高效新闻视频单元切分方法。该方法首先检测出新闻视频的镜头边界;然后从每个镜头中提取出关键帧,并计算出关键帧的直方图和SIFT特征;最后通过关键帧聚类获取新闻视频中的所有口播镜头,并以此为依据将新闻视频分割成多个语义单元。基于以上方法,开发了用于新闻视频单元切分的软件系统。该系统能够准确、高效地实现新闻单元的自动切分,有效地减轻视频切分时的工作强度,满足新媒体时代节目快速制作的要求。     

一种改进的新闻主持人镜头聚类算法

为解决传统新闻主持人聚类中背景变化与主持人位置大小变化影响聚类结果的问题,提出一种扩展人脸区域的主持人镜头模糊直方图聚类算法。引入扩展人脸区域（EFR）代替整个视频帧,采用模糊直方图（FCH）法对扩展人脸区域（EFR）进行聚类,与传统聚类方法比较,该方法有效地减少了背景对聚类结果的影响。采用背景变化剧烈的新闻视频为测试对象,经过实验验证此方法既有效的解决了背景变化带来的问题,又提高了查全率与准确率。     

基于音视频模板匹配的新闻视频识别方法

本文提出了一种基于音视模板匹配的新闻视频识别方法。在模板建立过程中,从新闻视频片头中的主题音乐提取音频模板,从主持人镜头中的扩展人脸区域提取视觉模板,这两者共同构成音视模板;在识别过程中,对电视视频流先进行音频模板匹配,然后由匹配通过的候选时间点定位到相应的视频镜头,接着通过视觉模板对镜头中的扩展人脸区域进行匹配,进而确定主持人镜头,最后完成新闻视频识别。实验结果表明,该方法计算效率高、简单易操作,具有较好的实用价值。     

多特征融合的新闻视频单元高效分割方法研究

为了提高新闻视频单元分割的准确性和稳定性,提出了多特征融合的新闻视频单元高效分割技术.首先,通过基于重要性区域分析的局部直方图比较方法,实现新闻镜头的准确检测,并提取出每个镜头的关键帧.然后,检测出关键帧中的人脸信息,并针对局部直方图和人脸特征实现关键帧聚类,从而确定口播镜头的位置.最后,通过解析串联单进一步优化新闻单元分割的结果.在省、市、县播出的多个新闻视频上进行实验,结果表明该方法具有较高的可靠性和稳定性,能够准确高效地实现新闻单元的自动分割.     

基于类内和类间损失的场景构造方法

为了便于在不同结构层次上对视频进行检索和浏览。可以把视频序列分为不同层次的逻辑单元。逻辑单元的层次由上到下可以分成序列、场景、镜头和帧。其中场景是时间上具有一定顺序关系的相似镜头的集合。文章提出了基于类内和类间损失的场景构造算法。首先利用时间约束和颜色直方图求得镜头距离；然后基于类内和类间损失对相似镜头进行聚类，得到镜头类；最后在分析镜头类的基础上构造场景。实验证明，构造的场景比较好的反映了视频的内容。     

一种基于多尺度实例学习的视频事件检测算法

现有的大多数视频事件检测方法首先从视频帧或视频快照中提取特征,然后对特征进行量化和汇集,进而为整个视频生成一个向量表示.最后的汇集步骤虽然简单高效,但是可能丢失时间局部信息,而这些信息对于确定长视频中事件发生的位置具有重要作用,从而削弱了事件检测的准确性.为此,本文首先将每个视频表示为多个“实例”,并将其定义为不同时间间隔的视频段.然后,针对每个视频的正实例比例已知和未知两种情况,提出基于多尺度实例学习的检测算法,在将实例标签看成隐藏潜在变量的同时推断出实例标签以及实例尺度的事件检测模型.最后,利用大规模视频事件数据集进行了全面的仿真实验,结果证明了本文算法具有显著的性能提升.此外,算法还可以确定视频中导致正检测的时间段的位置,进而对检测结果做出解释.     

基于模糊粗糙集的新闻视频镜头边界检测方法

镜头边界检测是实现基于内容的视频检索的一个重要步骤.为了将视频分割成镜头,现有的方法大都是首先提取大量的特征然后构造相异性测度函数.然而,太多的特征就会降低算法的效率.因此,有必要对镜头边界检测的规则进行特征约简.本文将粗糙集中的属性重要性和模糊粗糙集中的分类精度相结合定义了模糊粗糙算子,并构造了相异度检测函数.最后给出了镜头边界检测的一般性规则.由于本文检测方案的自适应性,因此适合于各种类型的新闻视频.用来自中央电视台的3个多小时的新闻视频所做的镜头边界检测实验获得了95.4%的查全率和96.1%的准确率.     

Content browsing and semantic context viewing through JPEG 2000-based scalable video summary

The paper presents a novel method and software platform for remote and interactive browsing of a summary of long video sequences as well as revealing the semantic links between shots and scenes in their temporal context. The solution is based on interactive navigation in a scalable mega image resulting from a JPEG 2000 coded key-frame-based video summary. Each key-frame could represent an automatically detected shot, event or scene, which is then properly annotated using some semi-automatic tools or learning methods. The presented system is compliant with the new JPEG 2000 Part 9 'JPIP - JPEG 2000 interactivity, API and protocols', which lends itself to working under varying transmission channel conditions such as GPRS or 3G wireless networks. While keeping the advantages of a single 2D video summary, like the limited storage cost, the flexibility offered by JPEG 2000 allows the application to highlight interactively key-frames corresponding to the desired content first within a low-quality and low-resolution version of the full video summary. It then offers fine grain scalability for a user to navigate and zoom into particular scenes or events represented by the key-frames. This possibility of visualising key-frames of interest and playing back the corresponding video shots within the context of the whole sequence (e.g. an episode of a media file) enables the user to understand the temporal relations between semantically related events/actions/physical settings, providing a new way to present and search for contents in video sequences.     

A MPEG video structure analysis scheme and its application to hierarchical video browser

This paper presents a novel scheme that automatically extracts out the hierarchical structure of MPEG video by detecting cuts and scenes, and a very userfriendly hierarchical video browser. The proposed scheme first conducts a fast, robust cut detection, chooses the representative frames of shots through clustering, calculates the connectivity between shots based on the similarities of the shots including a certain number of neighbors by using fuzzy reasoning, and extracts scenes by examining the connectivity change. The scheme is overall robust against threshold setting, cut missdetection and overdetection. We applied our scheme to parts of movies, the cut and scene detection rates are respectively higher than 95% and 75%. Considering the scheme is of high speed and in no need of knowledge on specific kinds of videos, these detection rates are quite satisfactory. The scheme is also feasible for other kinds of videos if they possess the feature that scene consists of a group of similar shots.     

新闻视频片花检索新方法

针对新闻视频中精品小栏目的点播需要靠人工剪辑,费时费力,提出了一种基于时空切片的片花检索方法。该方法根据片花的特点过滤掉大部分不相关的镜头,然后通过镜头合并规则得到候选片段,最后将候选片段融合为时空切片,并在候选片段时空切片的基础上采用分块主色调特征和纹理特征相结合的方法得到最后的相似片段。实验从查全率和定位精度两方面证明了该方法的有效性。     

一种检测足球视频中射门镜头的方法

体育视频中事件的检测是基于内容的体育视频检索的前提.根据足球比赛视频的特点,提出了一种有效的射门镜头检测方法,即利用主颜色区域检测和投影法来检测射门镜头,从而实现了按射门镜头对视频进行检索.最后给出实验结果及分析.     

Dominant sets based movie scene detection

Multimedia indexing and retrieval has become a challenging topic in organizing huge amount of multimedia data. This problem is not a trivial task for large visual databases; hence, segmentation into low- and high-level temporal video segments might improve the realization of this task. In this paper, we introduce a weighted undirected graph-based movie scene detection approach to detect semantically meaningful temporal video segments. The method is based on the idea of finding the dominant scene of the video according to the selected low-level feature. The proposed method starts from obtaining the most reliable solution first and exploit each solution in the subsequent steps recursively. The dominant movie scene boundary, which can be the highest probability to be the correct one, is determined and this scene boundary information is also exploited in the subsequent steps. We handle two partitioning strategies to determine the boundaries of the remaining scenes. One is a tree-based strategy and the other is an order-based strategy. The proposed dominant sets based movie scene detection method is compared with the graph-based video scene detection methods presented in literature.     

压缩域中基于支持向量机的镜头边界检测算法

针对如何进一步提高镜头边界检测精度问题,本文提出了一个基于支持向量机SVM (Support Vector Machine)的镜头边界检测算法.该算法利用视频压缩域中特征,如宏块类型,帧间对应宏块DC系数差和帧类型将视频帧分为发生切变的帧、发生渐变的帧和非镜头变换帧三类,从而实现视频的镜头分割.实验结果表明该算法对摄像机的运动和大物体的进入具有很好的鲁棒性,且没有大多数算法中阈值选择的困难,将我们的算法与2001 TREC评估中最佳指标进行了比较,在综合度量查全率和查准率的性能指标F1上,比2001 TREC评估中最佳指标高约8％.     

Video scene clustering by graph partitioning

A new video scene clustering method using graph partitioning is proposed. In the method, the shots of a video are grouped into clusters of similar scenes based on shot colour attributes; the number of scene clusters is not required to be known a priori. Experimental results are presented to show the effectiveness of the proposed method.     

A multi‐attribute shot segmentation algorithm for video programs

This paper describes a new algorithm for detecting cuts, thereby segmenting a video into shots. Our Webbased video library contains a large volume of news and documentary material; most of the transitions between shots in that type of programming are cuts, rather than dissolves or other complex transitions. We have developed an accurate multiattribute algorithm for detecting cuts in video programs. The algorithm uses a motion metric to identify a set of cuts, then uses luminance histograms to eliminate false cuts. Our experimental results show that this algorithm is more accurate than previous motionbased transition detection algorithms.     

电视编辑技术中的10点精要

视频编辑的基本要求是镜头与镜头之间的接点(编辑点)接得准确、平稳、自然，除对画面连贯性掌握之外，还应对镜头长短、镜头的使用原因、场景转换、节奏的掌握等作出正确的判断。主要介绍了10个常用的电视编辑技术要点，这些要点在电视编辑中是基础的，也是容易忽视的。     

一种新的动态视频摘要生成方法

本文提出了一种概要序列形式的视频摘要生成方法.在视频结构化的基础上,从减少信息冗余和保持本质内容两个角度出发,分别定义了镜头(场景)的冗余度和语义重要度,据此推导出视频内容缩减和保持的原则;之后,综合可理解性的要求,给出了一种自上而下的视频摘要生成方法.以电影动作片为例的实验结果证实了方法的有效性.