基于支持向量机的视频语义场景分割算法研究

Research on the method of video semantic scene constructing based on SVM

针对视频分割中存在的低层特征与高层语义之间"语义鸿沟"问题,在对视频进行镜头边界检测的基础上,引入视频语义概念矢量的定义,实现了一种基于支持向量机的视频语义场景构造方法。根据镜头关键帧画面语义的不同,提取镜头关键帧的颜色特征,并将其归一化;然后利用支持向量机对归一化后的特征量进行语义分类预测,从而生成语义矢量;将生成的语义矢量应用于已有的重叠镜头链方法,对镜头关键帧进行聚类,按语义差别构造出不同场景。实验结果证明了该方法的有效性。

Aimed at the problem of ＂semantic gap＂existing in video segmentation between low-level characteristic and high-level characteristic,the definition of video semantic concept vector was introduced.Based on the video shot boundary detection,a method of semantic scene construction based on support vector machine was implemented.First,according the difference of semantic in key frames,color characteristic was extracted and normalized.Then the characteristic was classified and predicted to generate semantic vector using support vector machine.Last,based on the generated semantic vector,＂overlapped shot linked＂method was employed to gather keyframe and construct different semantic scenes.The results of experiments show the effectiveness of this method.