在线聚类算法用于基于内容的镜头检索

视频数据具有一定的隐舍层次性,因此在对视频进行镜头分割后,可以利用模糊聚类算法依据相似度提取关键帧和关键镜头,对视频内容进行抽象概括,并以此时镜头进行检索.本文用非监督学习方法中的在线聚类算法自动提取镜头的关键帧,以相似方法对关键帧进行聚类以自动提取关键镜头,并对分类结果进行自我调整.将上述方法实现并用于镜头检索,获得了良好的检索结果,并减少了经验对聚类的影响,而且较好地表示了视频内容的层次性.

Content-Based Shot Retrieval by On-Line Clustering Algorithm

The video data have connotative hieratrchical structure. After the video is segmented by the shot, the fuzzy clustering approach can be used to extract key frames and key shots by the similarity measures, and then abstract the video content. The shot retrieval can be performed using this method. By using on-line unsupervised clustering, key frames can be automatically chosen. Then in the same way, key shots can be acquired from the frames with an appropriate self-adjusting method. Experimental results show superior performance and representation capability of the proposed approach for the video with less manual operation.