跨模态检索的鉴别子空间学习方法

Discriminative Subspace Learning Method for Cross-Modal Retrieval

许多现有的基于子空间学习的跨模态检索方法只集中于学习一个潜在的子空间,忽略了对鉴别性信息的充分利用,没有很好地保留语义结构信息。为了弥补这一不足,提出了一种跨模态检索的鉴别子空间学习方法(DSL),首先构建一个共享语义图来保留每个模态中的语义结构,随后引入希尔伯特-施密特独立性准则(HSIC)来保持样本的特征和语义相似度之间的一致性,最后构建角度重构方案。由此DSL可以弥补鉴别性数据使用不足的缺陷,更好地保留每个模态的语义结构信息。在两个常用的基准数据集上进行实验,结果表明上述方法相对于经典子空间学习方法更具有效性。

Many existing cross-modal retrieval methods based on subspace learning only focus on learning a latent subspace.They ignore the full use of discriminative information so that the semantic structure information is not well preserved.This paper proposes a discriminative subspace learning for cross-modalretrieval(DSL).Based on the DSL,a shared semantic graph was constructed to preserve the semantic structure within each modality.Subsequently,the Hilbert-Schmidt Independence Criteria(HSIC)was introduced to preserve the consistence between feature-similar⁃ityand semantic-similarity of samples.Thirdly,an angular reconstructive scheme was ntroduced,thus DSL can com⁃pensate for the shortcomings of insufficient use of discriminative data and better preserve the semantically structural information within each modality.The experimental results obtained on two well-known benchmark datasets demon⁃strate the effectiveness of the proposed method against the compared classic subspace learning approaches.