Proposed a novel method based on relevance vector machine for extracting the semantic information from keyframes of video, which was that HSV color histogram, MPEG-7 edge histogram and gray level cooccurrence matrix were combined for constructing feature vector set. Then it optimized the multi-classification classifier structure of binary tree based on graph partition model, applied the optimal binary tree for creating semantic multi-classification model, and used the active training strategy to optimize the training. Utilized RVM model to train and classify feature set of the keyffames for obtaining the semantic information eventually. The results based on lots of comparison experiments prove that the proposed method not only has excellent extraction rate, but also has better performance in time and model's sparsity when doing classification, compared with other methods.%将相关向量机理论应用于视频关键帧语义提取.该方法把关键帧中的HSV颜色直方图、MPEG-7边缘直方图和灰度共生矩阵相结合,建立特征标定向量集;基于图分割模型对二叉树多分类器结构进行优化,构建最优二叉树语义多分类模型并采用主动训练策略进行训练优化;利用RVM模型对关键帧特征向量集进行训练和检测,进而得到语义.实验结果表明,所提方法与其他方法相比,不但有较高的准确率,而且在模型的稀疏性、分类检测时间等性能方面也有很好的表现.
展开▼
