足球视频的语义颜色提取与语义镜头分割

Semantic color extraction and semantic shot segmentation for soccer video

在镜头颜色特征分析的基础上对足球视频的镜头进行了语义意义下的分割,即语义镜头分割.常用的基于主颜色的镜头分类方法只提取一种颜色作为分类特征,不能有效地处理语义颜色丰富的体育视频.将主颜色扩展为多个语义颜色,定义了颜色比例特征,再利用SVM对镜头进行分类,实验结果表明,颜色比例特征能够有效地提高镜头分类精度.考虑到视频语义颜色随时间和环境会发生变化,还给出了一种自适应的视频语义颜色提取算法,可以使语义颜色随环境的改变而自适应调整.

Based on the color feature analysis,this paper considers the shot segmentation issue from the perspective of semantic analysis,i.e.,the semantic shot segmentation.In most prior works they segment and classify the shot by using the dominant color of the field for soccer video,which can not properly utilize the plenty of color information on sports video.In this paper,we expand one dominant color in to several semantic colors,and define the Color Ratio Feature（CRF）.Then,the Support Vector Machine（SVM） framework is used for shot classification.The precision of shot classification shows that the color ratio feature is useful to improve the performance.Further,considering the temporal variations in the semantic color due to environment,this paper gives an adaptive semantic color extraction algorithm,and we evaluate the influence of classification precision on the number of semantic colors.