新媒体背景下的视频广告智能识别研究  

Research on intelligent identification of video advertisements in the context of new media

作　　者：白皓 BAI Hao(Shanghai Jian Qiao University,Shanghai 201306,China)

机构地区：[1]上海建桥学院,上海201306

出　　处：《电子设计工程》2021年第15期184-188,193,共6页Electronic Design Engineering

摘　　要：为提升新媒体背景下视频广告的识别精度与效率,研究新媒体背景下的视频广告智能识别方法。通过视频匹配算法分割目标视频镜头,获得数个与模式视频广告相匹配的视频段,采用3D卷积神经网络提取分割后所获取视频段的视频帧特征,计算特征的相似度,依据计算的相似度提取出各视频段中关键帧特征,将该特征当作数个当下特征块,通过多级连续排除算法以逐层漏斗型方式,排除掉与当下特征块无法匹配的候选块,逐层降低匹配块的总数,最终获取最优匹配块,即为所识别出的视频广告,实现新媒体背景下的视频广告智能识别。实验结果表明,该方法能够从视频数据库中识别出视频广告,具有较高的识别精度与识别效率,识别性能平稳优越。In order to improve the recognition accuracy and efficiency of video advertising in the context of new media,the intelligent recognition method of video advertising in the context of new media is studied.Through the video matching algorithm to segment the target video shot,several video segments matching the pattern video advertisement are obtained.The 3D convolution neural network is used to extract the video frame features of the segmented video segments,and the similarity of the operation features is calculated.According to the similarity after the operation,the key frame features in each video segment are extracted,and the features are treated as several current feature blocks,and the multi-level continuous arrangement is used In addition,the algorithm eliminates the candidate blocks that can't match the current feature blocks in a funnel-shaped way layer by layer,reduces the total number of block matching layer by layer,and finally obtains the best matching block,which is the identified video advertisement,so as to realize the intelligent recognition of video advertisement in the context of new media.The experimental results show that this method can recognize video advertisements from video database,which has high recognition accuracy and efficiency,and the recognition performance is stable and superior.

关 键 词：新媒体 视频广告 镜头分割 卷积神经 关键帧 匹配 

分 类 号：TN01[电子电信—物理电子学]