Collective Reconstructive Embeddings for Cross-modal Hashing.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Existing cross-modal hashing methods often overlook modality specifics, leading to information loss.
• Multi-modal integration complexity is a significant challenge in cross-modal retrieval.

Purpose of the Study:

• To propose Collective Reconstructive Embeddings (CRE), a novel approach for cross-modal hashing.
• To address both data heterogeneity and integration complexity in multimodal data.

Main Methods:

• Utilizing modality-specific models: cosine similarity for text and Euclidean distance for images.
• Unifying projections into a common reconstructive embedding in Hamming space.
• Incorporating code balance and uncorrelation criteria with an efficient iterative optimization algorithm.

Main Results:

• CRE demonstrates superior performance on challenging cross-modal retrieval tasks.
• Achieved state-of-the-art results across four widely-used multimodal benchmarks.
• Effectively preserves inter-modal similarity while reducing optimization complexity.

Conclusions:

• Collective Reconstructive Embeddings (CRE) offers a robust solution for cross-modal retrieval.
• The proposed method effectively handles heterogeneous data and complex integration.
• CRE advances the field of hashing-based approximate nearest neighbor search for multimodal data.