Hetero-Manifold Regularisation for Cross-Modal Hashing.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Cross-modal search is challenging due to data heterogeneity and integration complexity.
• Existing methods struggle with efficiently searching across different data modalities.

Purpose of the Study:

• To propose a novel method, hetero-manifold regularization (HMR), for efficient cross-modal search.
• To address the challenges of data integration complexity and heterogeneity in multi-modal data.

Main Methods:

• HMR learns hash functions by defining a hetero-manifold integrating multiple homogeneous data sub-manifolds.
• Similarity is measured using three-order random walks on the hetero-manifold.
• A cumulative distance inequality is introduced to manage discrete hash codes, transforming the problem into hetero-manifold regularized support vector learning.

Main Results:

• HMR significantly improves cross-modal search performance by combining hetero-manifold information and support vector machine generalization.
• Experiments demonstrate HMR's advantage over state-of-the-art methods on several cross-modal tasks.

Conclusions:

• HMR offers an effective solution for efficient cross-modal search with heterogeneous data.
• The proposed method shows superior performance and robustness in challenging cross-modal scenarios.