基于内容的视频片段检索技术研究

发布时间：2018-09-09 14:19

【摘要】： 近年来，随着视频信息资源的日益丰富，从大量的视频数据中检索感兴趣的视频片段已经成为目前视频检索研究的热点。人们期望在不久的将来像文本检索那样，由用户向系统提交查询的视频内容，，系统自动地返回符合查询需要的结果。在这种背景下，基于内容的视频检索应运而生，它综合了图像处理、模式识别、计算机视觉、图像理解等领域的知识，具有广阔的应用前景。本文以广告类视频、球拍类体育比赛视频和新闻视频为研究对象，主要研究了这三类视频的片段检索方法。提出了一种新的广告类视频片段检索方法。在定义查询片段镜头的相似镜头集的基础上，给出了两个片段一一对应相似镜头数目的计算方法，以排除伪相似片段，进而定义了片段的匹配函数；采用滑动镜头窗从连续的视频流中自动分割出多个相似片段；将相似镜头集映射为相似镜头矩阵，利用矩阵的特性来计算影响片段相似程度的各个因子，建立了相似片段的排序模型。与同样功能的其他方法相比，在确保较高的检索精度的前提下，提高了检索速度。融合视觉特征、运动特征和音频特征，提出了一种球拍类体育比赛视频的Rally精彩片段检索方法。提出了一种基于运动特征和主颜色特征的自动选择Global View Shots类的参考关键帧的方法，以便将镜头划分为Global View Shots类和Non-Global View Shots类；SVM用于识别镜头的重要音频类型；将音频信息应用到镜头类，建立了感兴趣事件检测的相应规则；利用回放的Rally片段和Rally精彩程度模型，从球拍类比赛中检索出符合多数人心理的Rally精彩片段。提出了一种基于视觉特征和语义信息的新闻视频片段检索方法。提出了一种新闻视频主题字幕文本检测方法，进而讨论了一种通过主题字幕文本和静音片段进行新闻视频故事分段的方法；基于低层视觉特征和主题字幕文本获得的语义信息建立了新闻片段的相似度模型，并应用相关反馈技术从新闻节目中检索出最符合用户需求的新闻故事片段。设计并初步实现了一个基于内容的视频片段检索实验平台，验证了上述片段检索方法的性能。
[Abstract]:In recent years, with the increasing enrichment of video information resources, retrieval of interested video fragments from a large number of video data has become the focus of video retrieval research. It is expected that in the near future, like text retrieval, users will submit the video content of the query to the system, and the system will automatically return the results that meet the needs of the query. In this context, content-based video retrieval emerges as the times require. It integrates the knowledge of image processing, pattern recognition, computer vision, image understanding and so on, and has a broad application prospect. In this paper, the video of advertising, racket sports and news video are taken as the research objects, and the retrieval methods of these three kinds of video fragments are mainly studied. In this paper, a new method of video segment retrieval for advertisement is proposed. Based on the definition of the similar shot set of the query segment shot, the method of calculating the number of similar shots corresponding to the two fragments one by one is given to eliminate the pseudo-similar fragments, and then the matching function of the fragments is defined. The sliding lens window is used to automatically segment several similar fragments from the continuous video stream, and the similar shot set is mapped to the similar lens matrix, and the factors affecting the similarity degree of the segment are calculated by using the properties of the matrix. The sorting model of similar fragments is established. Compared with other methods with the same function, the retrieval speed is improved on the premise of higher retrieval accuracy. Based on visual features, motion features and audio features, a Rally video retrieval method for racket sports competition video is proposed. A method of automatically selecting the reference key frame of Global View Shots class based on motion feature and main color feature is proposed in order to divide the shot into Global View Shots class and Non-Global View Shots class to recognize the important audio type of shot. The audio information is applied to the lens class, and the corresponding rules for detecting interesting events are established. By using the playback Rally fragment and the Rally wonderful degree model, the Rally wonderful fragments which accord with the majority of people's psychology are retrieved from the racket game. This paper presents a method of news video segment retrieval based on visual features and semantic information. In this paper, a method of text detection for news video topic subtitles is proposed, and then a method for news video story segmentation through topic caption text and mute segment is discussed. The similarity model of news segments is established based on the semantic information obtained from low-level visual features and subject subtitles. The relevant feedback technique is applied to retrieve the news story fragments that are most suitable for users' needs. A content-based video segment retrieval experimental platform is designed and implemented, and the performance of the above method is verified.
【学位授予单位】：南京理工大学
【学位级别】：博士
【学位授予年份】：2007
【分类号】：TP391.3

【相似文献】