In order to solve the problem that the graph-based serai-supervised learning methods neglect the video consistency in multimedia research area,this paper presented a new method video semantic annotation based on correlative kernel linear neighborhood propagation. The algorithm firstly structured the coefficient by kernel function, and through the coefficient to getting the samples that representative the low feature space. And then according to video correlation modeling, it structured the table between the semantic concepts. Finally, it completed the construction of graph, and used the video information that have been annotated spread to the video that not annotated. After that, the video annotation finished. The experimental results validate the effectiveness and superiority of the proposed method, the process of the video annotation addresses the insufficiency of labeled videos, improves the precision of annotation.%针对基于图的半监督学习方法在多媒体研究应用中忽略视频相关性的问题,提出了一种基于相关核映射线性近邻传播的视频标注算法.该算法首先通过核函数按照半监督学习调整后的距离计算出迭代标记传播系数；其次利用传播系数求得表示低层特征空间的样本,再根据视频相关性建模构造出语义概念间的关联表；最后完成近邻图的构造,并利用已标注视频信息迭代传播到未标注视频中,完成视频标注.实验结果表明,该算法不仅可以提高视频标注的准确度,还能弥补已标注视频数据数量的不足.
展开▼
