Please use this identifier to cite or link to this item: https://hdl.handle.net/10356/60868
Title: Mobile product recognition and recommendation
Authors: Hu, Chaoran
Keywords: DRNTU::Engineering
Issue Date: 2014
Abstract: With the rapid development of mobile phone and image processing technology, Mobile Visual Search has become a popular topic in recent years. Several innovative Mobile Visual Search applications have been released. However, some applications and algorithms still have the potential to be improved in accuracy and speed performance. For example, as a typical image recognition system, Vocabulary Tree (VT) Based Image Recognition has fine recognition accuracy but it takes a long time to query one image. The third process, Geometric Verification (GV) is time-consuming and therefore results in a slow recognition speed. In this Final Year Project, Fast Geometric Re-Ranking (FGRR) is developed and incorporated into the current VT Based Image Recognition system. It aims to speed up the overall image recognition and maintain or even improve the recognition accuracy. Experiments results show that FGRR achieves these objectives. FGRR alone improves VT recognition accuracy by around 1% and incorporated FGRR has maximum 0.8% accuracy improvement on VT Based Image Recognition. Besides, FGRR only takes a small portion of overall time consumption. When FGRR is incorporated to maintain same accuracy level, overall time consumption is reduced by 33%. To sum up, objectives of FGRR are achieved in this project.
URI: http://hdl.handle.net/10356/60868
Rights: Nanyang Technological University
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
Mobile Product Recognition and Recommendation.pdf
  Restricted Access
2.11 MBAdobe PDFView/Open

Page view(s) 50

125
checked on Oct 19, 2020

Download(s) 50

10
checked on Oct 19, 2020

Google ScholarTM

Check

Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.