Journal
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
Volume 63, Issue -, Pages -Publisher
ACADEMIC PRESS INC ELSEVIER SCIENCE
DOI: 10.1016/j.jvcir.2019.102577
Keywords
Hash function; Deep learning; Image retrieval; Convolutional neural network (CNN); Deep hashing
Funding
- National Natural Science Foundation of China [61872004, 61672035]
- Anhui Scientific Research Foundation for Returned Scholars
Ask authors/readers for more resources
Hashing is one of the most popular image retrieval technique since its fast-computational speed and low storage cost. Recently, deep hashing methods have greatly improved the image retrieval performance in contrast to traditional hashing method. However, the binary hashing representation is only generated from the global image region, which may result in sub-optimal hashing code. Inspired by the latest advance in spatial attention mechanism, we propose an novel end-to-end deep hashing framework which composes of two sub-networks. One sub-network uses spatial attention model to determine the local features from more specific region of interest, another sub-network extracts the global features from original image. By combining the local and global features with learnable hash functions, the proposed deep hashing framework can optimize the deep hash function and high-quality binary code jointly. Numerous experiments on two large scale image benchmarks datasets have shown that the proposed method is superior to other existing methods for image retrieval. (C) 2019 Elsevier Inc. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available