摘要： 目前大多数的视频语义概念提取研究没有考虑到视频多模态之间的关联共生特性，而在样本的标注方面采用自定义的概念进行标注，会影响语义概念提取的准确率。针对上述问题，提出结合Simfusion算法和用本体知识库标注样本的方法提取视频的语义概念，该方法根据镜头内容变化提取关键帧，在提取出镜头内容时，有效地利用镜头多模态之间的时序关联共生特性，同时运用本体知识库中的概念标注样本、训练分类器，弥补传统方法在标注样本时存在的主观、不规范等不足。实验结果表明，该方法在视频语义概念提取的研究中，有较高的准确度、可操作性强。

关键词: Simfusion算法, 本体知识库, 时序关联共生特性, 多模态, 视频语义概念

Abstract: Most of the researches in extracting semantic concepts do not consider the temporal associated co-occurrence characteristic of multimodes, and label the training set using self-define concepts, thus affecting the accuracy of semantic concepts extraction. Aiming at these problems, this paper brings forward a new approach based on Simfusion algorithm and labeling the training set using ontology repository to extract semantic concepts of video. The method extracts key-frame according to the content of the shots, and makes the most of temporal associated co-occurrence characteristic during in multimode. Meanwhile, the method labels the sample set using the ontology repository and training classifier, thus offsetting insufficiency in subjectivity, incorrect. Experimental result shows that the method can get a better accuracy, well operability and universality in the research of semantic concepts of video.

Key words: Simfusion algorithm, ontology repository, temporal associated co-occurrence characteristic, multimode, video semantic concepts

中图分类号: 

• TP391.4