基于双模态特征和支持向量机的视频自动分类算法研究

发布时间：2018-01-27 06:27

本文关键词： 视频分类特征双模态二次预测支持向量机　出处：《上海交通大学》2010年硕士论文　论文类型：学位论文

【摘要】： 视频内容的自动分类算法是计算机视觉领域中一个重要的研究课题,它为日益增加的视频数据的管理提供了方便,基于内容的视频自动分类作为视频传播控制的一类关键技术在对网络媒体进行有序管理的需求下至关重要。基于视频自动分类技术的应用,媒体网站可以把海量的视频内容进行自动分类,从而进行更有效的组织、存储和检索,还可实现对不良视频信息,如恐怖暴力视频的自动初步筛选。视频自动分类算法的性能极大地依赖于视频特征的提取和分类算法的选取,本文从对视频内容和视频风格类型的角度出发,提出了基于视觉和音频双模态特征组合的视频特征提取方法,和改进支持向量机(SVM)视频分类算法,实现了对卡通、广告、音乐、新闻、体育这五类常见视频的自动分类,以及对电影中恐怖暴力场景的自动识别。首先,在分析现有的视频分类算法的基础上,针对现有算法存在的问题,通过分析五类常见视频在视觉上的差异,本文提出了新的特征表达方案即MPEG-7视觉描述子组合模型,从颜色、纹理、形状、运动四个方面提取了共九种描述子来构成新的整体视觉特征,取得了较好的效果;在识别恐怖暴力场景时,本文根据这些场景的特点采用了视觉和音频两种模态的特征,相比单一特征增加了场景模式匹配的准确率,在有效性和区分度上达到了满意的效果。在选择并提取了合适的特征后,针对目前统计方法中存在的通过小样本集很难设计有效分类器的问题,本文提出了基于支持向量机的视频自动分类算法,并对分类器的判决策略方法进行了改进,提出了基于支持向量机1-1方法的二次预测机制,进一步提高了支持向量机多分类方法的准确率。仿真实验的结果表明:本文算法在特征选择方面突出了不同类别视频的差异性,增强了待分视频的区分能力;其次,改进的二次预测机制提高了支持向量机的多视频分类的性能;最后,与目前的相关类似算法进行了对比实验,五类视频的分类实验和恐怖暴力场景的识别实验均证明了本文算法在视频分类准确率方面的优越性能。
[Abstract]:The automatic classification algorithm of video content is an important research topic in the field of computer vision. It provides convenience for the increasing management of video data. As a kind of key technology of video propagation control, content-based video automatic classification is very important under the demand of orderly management of network media. Media websites can automatically classify large amounts of video content, so as to organize, store and retrieve more effectively, and can also realize the automatic preliminary screening of bad video information, such as terrorist violence video. The performance of the automatic video classification algorithm is greatly dependent on the feature extraction and classification algorithm selection. This paper starts from the point of view of video content and video style types. This paper proposes a video feature extraction method based on the combination of visual and audio features, and an improved support vector machine (SVM) video classification algorithm to achieve cartoon, advertising, music, news. The automatic classification of the five common sports videos and the automatic recognition of the scenes of horror and violence in movies. First of all, based on the analysis of the existing video classification algorithms, aiming at the existing problems, through the analysis of five kinds of common video in the visual differences. In this paper, a new feature representation scheme, MPEG-7 visual description sub-combination model, is proposed. Nine descriptors are extracted from color, texture, shape and motion to form a new overall visual feature. Good results have been achieved; According to the characteristics of these scenes, this paper adopts the features of visual and audio modes, which increases the accuracy of scene pattern matching compared with a single feature. Satisfactory results have been achieved in terms of effectiveness and differentiation. After selecting and extracting the appropriate features, aiming at the problem that it is difficult to design effective classifier through small sample set in the current statistical methods, this paper proposes an automatic video classification algorithm based on support vector machine. The decision strategy method of classifier is improved, and a quadratic prediction mechanism based on support vector machine (SVM) 1-1 method is proposed, which further improves the accuracy of SVM multi-classification method. The simulation results show that: the algorithm in feature selection highlights the differences of different types of video, and enhances the ability to distinguish the video to be divided; Secondly, the improved quadratic prediction mechanism improves the performance of support vector machine (SVM) multi-video classification. Finally, compared with the current similar algorithms, five kinds of video classification experiments and terrorist violence scene recognition experiments have proved the superiority of this algorithm in the accuracy of video classification.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：TP391.41

【引证文献】