Deep multi-similarity hashing technique based on multi-label semantics

Abstract A deep multi-similarity hashing technique based on multi-label semantics has been proposed in this paper as a means of a new controlled deep hashing system for multi-label image retrieval. Our proposed methodology comprises hash code learning and semantically-aware similarity matrix creatio...

Full description

Saved in:
Bibliographic Details
Published inJournal of physics. Conference series Vol. 2506; no. 1; pp. 12001 - 12005
Main Authors Cui, Lichao, Bi, Yuyang, Liu, Mingxin
Format Journal Article
LanguageEnglish
Published Bristol IOP Publishing 01.05.2023
Subjects
Binary codes
Labels
Learning
Physics
Retrieval
Semantics
Similarity
Online AccessGet full text

Cover

More Information
Summary:Abstract A deep multi-similarity hashing technique based on multi-label semantics has been proposed in this paper as a means of a new controlled deep hashing system for multi-label image retrieval. Our proposed methodology comprises hash code learning and semantically-aware similarity matrix creation. To create a similarity matrix that considers semantic context, we integrate label-level and semantic-level similarity. Using the higher-order statistics of deep features as inputs, we meticulously craft the multi-similarity loss and quantization error loss for hash code learning, ensuring that the learned binary codes retain their high-ranking similarity. Our suggested strategy is successful, as shown by several testing on CIFAR-10 and NUS-WIDE.
ISSN:1742-6588
1742-6596
DOI:10.1088/1742-6596/2506/1/012001