Title: Spatial descriptor embedding for near-duplicate image retrieval

Authors: Yunlong Wang; Zhili Zhou

Addresses: Jiangsu Engineering Center of Network Monitoring and School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing, China ' Jiangsu Engineering Center of Network Monitoring and School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing, China; Department of Electrical and Computer Engineering, University of Windsor, Windsor, Ontario, Canada

Abstract: The existing methods for near-duplicate image retrieval are mostly dependent on the bag-of-words (BOW) model. However, the procedure of quantisation to the low discrimination of visual words causes many false local matches. In this paper, we propose a novel spatial descriptor embedding method for near-duplicate image retrieval, which encodes the relationship of the SIFT dominant orientation and the exact spatial position between local features and their context to be spatial descriptors, and then embeds them in the index to improve the distinctiveness of visual words. Moreover, a secondary matching structure for spatial descriptors matching is used to effectively and efficiently implement the near-duplicate image retrieval. Experimental results on Copydays illustrate that our method achieves superior performance to the state of art methods.

Keywords: bag-of-words; BOW; spatial descriptor embedding; image retrieval; image search; near-duplicate image retrieval; partial-duplicate image retrieval; image copy detection; bag-of-visual-words; local feature; local descriptor; SIFT.

DOI: 10.1504/IJES.2018.091787

International Journal of Embedded Systems, 2018 Vol.10 No.3, pp.241 - 247

Received: 19 Jul 2016
Accepted: 15 Jan 2017
Published online: 16 May 2018 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article