A New VLAD Method with Dense SIFT Selection Application in Image Classification

DOI

10.2991/caai-17.2017.127How to use a DOI?

Keywords

image classification; features selection; SLIC; VLAD(key words)

Abstract

Since Dense SIFT causes a long time spending during clustering due to an excessive order of magnitude, and its feature descriptors reserve excessive insignificant features, we present a new method that using SLIC to select descriptors to address this problem. Furthermore, when VLAD aggregates, the partial directions of feature vectors have the excessive data offset and still distorts after the dimension deduction treatment. Regarding such issue, the algorithm that possesses the optimized clustering descriptor with feature membership information called FS-VLAD is proposed. The algorithm adopts the principle of the fuzzy cost function with the smallest deviation regarding the quadratic sum of the neighbor clustering center to calculate the feature membership degree. After conducting classification test, the result demonstrates that in comparison with the mainstream Dense SIFT + VLAD classification model, the new methods could improve by around 15%, and possesses better generality.

Copyright

© 2017, 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  - Zhi Qian
AU  - Qijun Hong
AU  - Gang Huang
AU  - Pingping Liu
AU  - Yuanjie Yan
AU  - Min Xie
PY  - 2017/06
DA  - 2017/06
TI  - A New VLAD Method with Dense SIFT Selection Application in Image Classification
BT  - Proceedings of the 2017 2nd International Conference on Control, Automation and Artificial Intelligence (CAAI 2017)
PB  - Atlantis Press
SP  - 568
EP  - 574
SN  - 1951-6851
UR  - https://doi.org/10.2991/caai-17.2017.127
DO  - 10.2991/caai-17.2017.127
ID  - Qian2017/06
ER  -