Logo Retrieval with Representation Error of Self-taught Encoding

DOI

10.2991/isci-15.2015.123How to use a DOI?

Keywords

Logo Retrieval; Representation Error; Feature Point clustering

Abstract

Logo retrieval in real-world scenarios has numerous potential applications in computer vision. Due to occlusion, illumination , non-rigid distortion and other reasons, the accuracy of feature matching in natural images is far lower than that in the print objects. For such a challenging task, a lot of papers have conducted very fruitful work. The algorithm finds approximate matching points in the images by locality sensitive hashing algorithm. Given matched points’ position information, matched points are divided into several groups. With RANSAC algorithm, each group of matched points are divided into inlier points set and outlier points set, the candidate windows of logos can be mapped. Finally by calculating the representation error score of overlapping candidate windows, the lower score regions are eliminated, and the higher score regions are remained. The result of experiment shows that our approach can effectively locate more than one logo areas in an image, improving the recall of retrieval. And it also improves the mean Average Precision scores greatly by sorted files with representation error score.

Copyright

© 2015, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Wei Liu
AU  - Yunxing Ruan
AU  - Xia Cai
PY  - 2015/01
DA  - 2015/01
TI  - Logo Retrieval with Representation Error of Self-taught Encoding
BT  - Proceedings of the 2015 International Symposium on Computers & Informatics
PB  - Atlantis Press
SP  - 927
EP  - 934
SN  - 2352-538X
UR  - https://doi.org/10.2991/isci-15.2015.123
DO  - 10.2991/isci-15.2015.123
ID  - Liu2015/01
ER  -