一种基于模糊信息粒化的视频时空显著单元提取方法

Using Fuzzy Information Granulation for Spatio-temporal Salient Unit Detection in Video Sequences

提出一种基于模糊信息粒化的视频时空显著单元提取方法,为视频分析及检索等高层应用提供一个有效的内容表示模式.本文首先提出了一种类相关的特征粒化方法,粒化后的模糊粒特征简化了分类关系且在一定程度上解决了感知主观性问题,因而通过简单的分类器可以有效地提取空域中具有高视觉感知显著度的区域(简称为显著区域);其次,通过对显著区域的时域一致性分析提取视频序列中时域连续的显著区域集合,定义为时空显著单元.提取的时空显著单元能作为一种较为通用的语义级内容表示模式.实验结果分别从时域和空域两个方面验证了本文方法的有效性.

In this paper, we propose an approach to automatic extraction of spatio-temporal salient unit from video sequences.Firsfly,a class-related feature granulation algorithm is proposed, which can map original feature space to concept space based on fuzzy information granular. To detect spatial salient regions, segmented homogenous regions are classified according their prominent importance. Then, salient regions are tracked and the results of tracking are sequences of temporal coherent regions, called spatio-temporal salient unit. Experimental results verify efficiency of proposed approach in spatial and temporal aspects.