论文部分内容阅读
作为视频检索的一种重要线索,音频检测和分类受到广泛关注并已成为一个热门的研究方向.在新闻视频先验模型和结构的基础上,提出一种基于选择性集成 SVM(SEN-SVM)的分类器设计方法.从而将新闻视频划分成静音、音乐、语音和带有背景音乐的语音这4种类型.用8514s 的真实新闻音频数据所作的仿真实验结果表明:所提出基于选择性集成 SVM 的新闻音频自动分类算法的平均准确率高达98.2%,远远高于单纯基于 SVM 的方法和传统的基于门限的方法.
As an important clue of video retrieval, audio detection and classification has drawn more and more attention and has become a hot research area.Based on the priori model and structure of news video, this paper proposes a new method based on SEN-SVM (Selective Ensemble SVM) The classification of news video is divided into 4 types: mute, music, speech and speech with background music.The simulation results using 8514s real news and audio data show that the proposed SVM based on Selective Integration SVM The average accuracy rate of news audio automatic classification algorithm is as high as 98.2%, which is much higher than pure SVM-based methods and traditional threshold-based methods.