基于多视角对称非负矩阵分解的跨模态信息检索方法

针对跨模态信息检索的策略和核心问题,从提升检索性能的角度,分析了多视角对称非负矩阵分解方法用于跨模态检索的优势,提出了一种新的基于对称非负矩阵分解的跨模态检索框架。首先在Wikipedia、Pascal公开数据集上习得一致的子空间表示;然后基于该子空间,设计了一种实时样本在子空间中的投影方法。与典型相关分析、语义匹配和偏最小二乘回归相比,在MAP和PR曲线这2个指标上,本文所提出的方法具有最优的性能表现,表明了该方法应用于跨模态信息检索任务中的潜力。

Cross-modal information retrieval method based on multi-view symmetric nonnegative matrix factorization

This article summarizes the strategies and core issues in cross-modal information retrieval and analyses the advantages of multi-view symmetric nonnegative matrix factorization for cross-modal retrieval in terms of improving retrieval effect. A new cross-modal retrieval framework based on symmetric non-negative matrix factorization is proposed. Firstly, a consistent subspace representation is learned from the Wikipedia and Pascal datasets. Then, based on the subspace, a method of mapping real-time samples into subspaces is designed. Compared with the canonical correlation analysis, semantic matching and partial least squares regression, the proposed method has the best performance in terms of MAP and PR curves. The results demonstrate that the proposed algorithm has the potential ability in the task of cross-modal information retrieval.