基于聚类算法的多特征融合关键帧提取技术研究

发布时间：2018-04-02 10:31

本文选题：关键帧提取　切入点：多特征融合　出处：《华中科技大学》2012年硕士论文

【摘要】：关键帧提取技术是基于内容的视频检索核心技术之一，对于内容复杂、场景繁多、动作丰富的视频类型（如动画片、广告片、动作片），现有的关键帧提取方法性能并不理想。第一，，关键帧集合代表性不够，不能有效代表原视频内容；第二，关键帧集合存在冗余，不够简洁。如果用这样的关键帧集合对原视频数据进行索引，那么视频检索的快速性和准确性将会受到很大的影响。为应对上述亟待解决的难点，有必要对视频序列关键帧提取方法进行创新；另一方面，由于视频文件数据量庞大，为了增强实用性，有必要提高关键帧提取系统的离线计算效率。本文针对内容复杂、场景繁多、动作丰富的视频类型，提出了一种新的基于聚类算法的多特征融合的关键帧提取方法。首先，将多种特征进行融合，再进行相似性度量可以更加完整全面的描述内容复杂的视频；其次利用聚类算法依据场景对视频序列进行聚类，免去了在场景繁多的情况下镜头检测分割的困难和繁杂；再次，依据运动量极小值标准来提取关键帧，能更准确的代表动作丰富的视频内容。同时本文对所采用的聚类算法进行了改进，提高了关键帧提取系统的离线计算效率。本文最后设计并实现了关键帧提取系统，通过对不同方法的提取结果进行定量分析与对比，表明本文提出的方法具有较高的性能和准确度。本文的研究对于促进关键帧提取技术和基于内容的视频检索的发展应用具有理论及应用价值。
[Abstract]:Key frame extraction is one of the core techniques of content-based video retrieval. The performance of the existing key frame extraction methods is not ideal. First, the key frame set is not representative enough to represent the original video content effectively; second, the key frame set has redundancy. If the original video data is indexed with such a set of key frames, the speed and accuracy of video retrieval will be greatly affected. In order to deal with the above difficulties, it is necessary to innovate the method of video sequence key frame extraction. On the other hand, because of the huge amount of video file data, in order to enhance the practicability, It is necessary to improve the off-line computing efficiency of the key frame extraction system. In this paper, a new key frame extraction method of multi-feature fusion based on clustering algorithm is proposed for the video types with complex content, multi-scene and rich action. Then the similarity measurement can more complete and comprehensive description of the complex content of video. Secondly, clustering algorithm is used to cluster the video sequences according to the scene, which avoids the difficulty and complexity of shot detection and segmentation in the case of a wide range of scenes. Thirdly, The key frame can be extracted according to the minimum motion quantity standard, which can represent the rich video content more accurately. At the same time, the clustering algorithm used in this paper is improved to improve the off-line computing efficiency of the key frame extraction system. Finally, a key frame extraction system is designed and implemented, and the results of different methods are quantitatively analyzed and compared. The results show that the proposed method has high performance and accuracy. The research in this paper has theoretical and practical value for the development and application of key frame extraction technology and content-based video retrieval.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41

【参考文献】