基于图稀疏Group Lasso算法的农业科教视频多语义标注方法

Agriculture Video Semantic Annotation Using Graph Sparse Group Lasso

针对农业领域的视频标签,多以人工方式标注不利于大量视频准确快速标注、检索的问题,提出了一种基于图稀疏Group Lasso模型的农业科教视频多语义标注方法:通过添加待测镜头与视频组间组内的稀疏约束,得到待测镜头在视频集内的稀疏编码,结合视频集内人工标注的标签,进行多语义的标注。农业科教视频多语义标注的试验表明,该方法能够实现语义的快速标注,并使得农业视频多语义标注的F综合指标达到64%。农业视频多语义标注效果,不仅可满足用户个性化的信息需求,同时也为农业知识视频检索等相关领域,提供了参考方案。

In agriculture education video analysis research area, manually semantic annotation requires tremendous human power. In order to provide an efficient and effective solution for semantic agriculture education video indexing and fast retrieval, in this paper, we propose a new video semantic video annotation scheme using graph sparse group lasso. With inter-group and intra-frame sparse constraints between the testing video shot and the annotated video group, a set of sparse reconstruction coefficients are estimated by solving a lasso optimization problem. And then multiplesemantic tags are annotated with the same coefficient. The experiment results on agriculture education video show that our proposed algorithm can achieve F-Measure to 64%. This new agriculture education video annotation algorithm can provide semantic information for retrieval.