Shared Predictive Cross-Modal Deep Quantization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• The increasing volume and diversity of data necessitate efficient cross-modal similarity search.
• Quantization-based methods show superior performance over hashing in single-modal search.

Purpose of the Study:

• To propose a deep quantization approach for efficient cross-modal similarity search.
• To leverage deep neural networks for quantization-based cross-modal similarity search.

Main Methods:

• Developed Shared Predictive Deep Quantization (SPDQ), a novel deep learning approach.
• Formulated shared and private subspaces across modalities, learning representations in a reproducing kernel Hilbert space.
• Employed label alignment and supervised quantization training to preserve semantic information and minimize quantization error.

Main Results:

• SPDQ effectively preserves both intramodal and intermodal similarities.
• The approach significantly reduces quantization error compared to existing methods.
• Experiments on benchmark datasets demonstrate superior performance over state-of-the-art techniques.

Conclusions:

• SPDQ offers an efficient and effective solution for cross-modal similarity search.
• The proposed deep quantization method advances the field of cross-modal retrieval.