基于声像特征的场景检测 *

视频的结构分析是实现视频基于内容组织和检索的基础。目前 ,已经有很多用于视频镜头分割的成熟算法 ,但准确探测视频场景边界还比较困难。提出了一种融合视频中音频与可视特征进行场景检测的方法。该方法首先分别依据镜头的声、像特征相关性来对镜头进行聚类 ,然后综合处理依声、像相关性得到的镜头聚类来获取场景。实验结果证明 ,此方法较一般使用单一特征的场景检测方法提高了探测的准确率 ,同时也降低了误判率。

Scene detection based on combined audio and visual feature

Video structure extraction is essential to content-based organization and retrieval of video. While many robust shot segmentation algorithms have been presented, it is still difficult to identify scene accurately. This paper presented a scheme for determining scene which clustered shots combining audio and visual features of video. In particular, this method first clustered shots with respect to each feature, and then determined scene integrating the different cluster. Results from experiments showed that this approach to be potent.