Dual Attention Triplet Hashing Network for Image Retrieval

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Learning-based hashing is efficient for large-scale image retrieval.
• Deep hashing methods often produce repetitive and correlated hash codes, limiting performance.

Purpose of the Study:

• To propose a novel deep hashing network, Dual Attention Triplet Hashing Network (DATH).
• To address limitations of existing methods by reducing redundant information in hash codes.
• To improve the efficiency and accuracy of large-scale image retrieval.

Main Methods:

• DATH utilizes a two-stream Convolutional Neural Network (ConvNet) architecture.
• One stream captures spatial semantic relevance, while the other captures channel semantic correlation.
• The network is optimized using a combination of triplet likelihood loss and classification loss, leveraging label information.

Main Results:

• DATH demonstrated state-of-the-art performance on benchmark datasets.
• The proposed dual-attention mechanism effectively reduces repetitive and correlated information in hash codes.
• The integrated loss functions enhance the utilization of label information for better network optimization.

Conclusions:

• DATH offers an effective solution for large-scale image retrieval by improving hash code quality.
• The dual-attention mechanism and combined loss functions represent a significant advancement in deep hashing techniques.
• The network achieves superior performance compared to existing methods on standard datasets.