提出了隐含语义相关性分析的处理方法,对描述视频内容结构的视频文档矩阵进行奇异值分解和变换,可得到表达可视特征与语义间隐含相关性的关系矩阵,用于视频检索。分析和实验表明,所提出的方法能保留视频内容结构中核心的语义元素,消除冗余的相关性干扰,还能够通过矩阵降维减少计算量,改善视频语义内容检索的效果。
A method of video features correlation analysis is proposed,by using the latent self-correlations and cross-correlation which exist in the visual features and semantics,it can retain the core element of the semantic structure in video content,and eliminate the interference induced by the redundant relationships.It is proved in practice that proposed method make better outcome of semantic-based video retrieval,and to improve the computational efficiency by reducing the size of matrix.