摘要： 提出了一种结合音视频双重特征检测视频内容的新方法，以提高对视频内容的识别准确率。该方法分别对视觉特征和音频特征进行分析，引入支持向量机对音频段进行分类，并综合音视域的分析结果对视频内容进行判断。针对特殊视频片断进行分析，证明结合音视特征的分析方法可行有效，可应用于视频内容监控及特定视频片段的检索与分割。

关键词: 音频分割, 肤色检测, 支持向量机, 视频内容识别

Abstract:

A novel audio-visual feature-based framework for porn video segment detection is presented, which extracts and analyses audio and video features independently, and combines their results to give out a final detection result. A support vector machine (SVM) is used to learn and then classify the audio sections, and skin info is used to give out a visual analysis result. Experiment results indicate that this approach has a high recognition rate, so it can be adopted to surveillance for video streams and segmentation of particular content in videos.

Key words: audio segmentation, skin detection, support vector machine (SVM), video content recognition

中图分类号: 

• TP18