摘要
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。
A news video recognition method is presented based on audio-video template matching in this paper. During the process of template build- ing, the audio template is extracted from the theme music of news video and the visual template is extracted from the extended face region of the anchor shot. During the process of recognition ,firstly, audio model matching is conducted upon news video stream to get candidate time points. Then video shots are located corresponding to these time points and match the extended face region detected in video shots with visual model. In the end, the anchorperson shots are fixed, thus the process of news video recognition is finished. Experimental resuhs show that this method is of low computing complexity, high detecting accuracy and fairly good practical value.
出处
《电视技术》
北大核心
2013年第23期238-240,共3页
Video Engineering
关键词
新闻视频
音视频模板
视频检索
news video
audio-video template
video retrieval
