Abstract:
This paper proposes a news story segmentation method based on Naïve Bayes model. It gets the candidate boundary points through shot detection, extracts multi-modal middle-level features at the boundary points such as visual, audio type, motion and caption to generate a feature set as the input of the model, and uses trained Naïve Bayes model to classify the candidate boundary points. After post-processing upon the result, the non-news parts are removed and the news parts are saved. Experimental results show that the method is effective and adaptive to the different kinds of news programs, and it achieves satisfactory precision and recall.
Key words:
story segmentation,
Naï,
ve Bayes model,
multi-modal fusion,
middle-level feature,
video retrieval
摘要: 提出一种基于朴素贝叶斯模型的新闻视频故事分割方法。通过对新闻视频进行镜头检测,获得候选故事边界点,从候选边界点周围镜头提取多模态中级特征,形成属性集合作为输入,应用朴素贝叶斯模型对候选边界点进行分类后对结果进行后处理,得到新闻故事。实验结果表明,该方法获得了较高的查准率和查全率,对不同类型的新闻节目有良好的适应性。
关键词:
故事分割,
朴素贝叶斯模型,
多模态融合,
中级特征,
视频检索
CLC Number:
PENG Tian-qiang; LI Bi-cheng. News Story Segmentation Method Based on Naïve Bayes Model[J]. Computer Engineering, 2009, 35(20): 178-180.
彭天强;李弼程. 基于朴素贝叶斯模型的新闻故事分割方法[J]. 计算机工程, 2009, 35(20): 178-180.