Deep Collaborative Multi-view Hashing for Large-scale Image Search.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• Hashing accelerates large-scale image search by mapping high-dimensional features to binary Hamming space.
• Multi-view hashing leverages multiple features for binary hash learning.
• Existing methods struggle with deep feature correlations or insufficient semantics.

Purpose of the Study:

• To propose a novel Deep Collaborative Multi-view Hashing (DCMVH) method.
• To deeply fuse multi-view features and collaboratively learn hash codes.
• To address limitations of shallow and unsupervised deep multi-view hashing.

Main Methods:

• DCMVH employs view-specific networks for hidden representation extraction.
• A fusion network learns multi-view fused hash codes.
• Instance-wise and pair-wise semantic labels enhance discriminative capabilities and feature complementarity.
• A fast discrete hash optimization method is utilized.

Main Results:

• DCMVH achieves substantial performance improvements over state-of-the-art methods.
• The framework effectively exploits the complementarity of different view features.
• Deep fusion enhances the discriminative capability of representation layers.

Conclusions:

• DCMVH offers a powerful deep learning framework for multi-view hashing.
• The method significantly advances the state-of-the-art in multi-view image search.
• Effective deep fusion and collaborative learning are key to improved performance.