Semi-Supervised Cross-Modal Retrieval Based on Discriminative Comapping

Figure 2

The flowchart of the proposed method. Images are represented by square icons, and the texts are represented by round icons. The images and texts with full lines are labelled data, and the rest are unlabelled data. Different colours indicate different categories. Our method adopts LDA to make different classes separate and make full use of the semi-supervised strategy to maximize the correlation between different modals simultaneously.