Please use this identifier to cite or link to this item: http://scholarbank.nus.edu.sg/handle/10635/139712
Title: CROSS-DOMAIN IMAGE RETRIEVAL WITH ATTENTION MODELING
Authors: JI XIN
Keywords: cross-domain, product search, feature learning, attention modeling, metric learning, convolutional neural networks
Issue Date: 29-Sep-2017
Source: JI XIN (2017-09-29). CROSS-DOMAIN IMAGE RETRIEVAL WITH ATTENTION MODELING. ScholarBank@NUS Repository.
Abstract: With the proliferation of e-commerce websites and the ubiquitousness of smartphones, cross-domain product image retrieval using images taken by smartphones as queries to search products on e-commerce websites is emerging as a popular application. Since images from two domains exhibit different characteristics, it is a challenge to gain discriminative feature representations which contain domain-specific features while preserving domain-invariant features. Another challenge is to locate the attention for both shop-domain images and user-domain images. Particularly, database images of the target products from the e-commerce websites are typically displayed with other accessories, while query images taken by users contain noisy background and large variations. In this paper, we exploit rich tag information available from e-commerce websites to locate the attention of database images and we use each candidate image in the database as the context to locate the query attention. We propose novel deep convolutional neural network architectures, namely TagYNet and CtxYNet, to learn the attention weights. Discriminative representations of the images are then extracted for retrieval based on the learned attention. Experimental results on two well-collected public datasets confirm that our approaches have significant improvement over the existing approaches in terms of the retrieval accuracy and efficiency.
URI: http://scholarbank.nus.edu.sg/handle/10635/139712
Appears in Collections:Master's Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
JiX.pdf3.18 MBAdobe PDF

OPEN

NoneView/Download

Page view(s)

4
checked on Apr 19, 2018

Download(s)

3
checked on Apr 19, 2018

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.