Generic Object Regions Matching Based VLAD Model for Image Retrieval

DOI

10.2991/mmebc-16.2016.273How to use a DOI?

Keywords

Image Retrieval, Multi-threshold Segmentation, Gaussian Mixture Model, VLAD, SURF

Abstract

In the popularly used BoVW and VLAD models for image retrieval, feature extraction is easily affected by the color and textures of images. Also, the clustering results of K-means algorithm used in these two models is usually affected by the initial cluster centroids. In order to solve these problems, a generic object regions matching based VLAD model for image retrieval is proposed. In this model, multi-threshold for image segmentation is proposed for the extraction of SURF features. Then the location information of SURF features are utilized for the Gaussian Mixture Model clustering instead of K-means algorithm. Finally, VLAD descriptors are calculated according to the features in each cluster and used for similar image searching. Our proposed Multi-Threshold Segmentation and SURF Feature Location based Clustering algorithm can improve the matching accuracy of features and obtain a better codebook such that the feature distribution can be better expressed. Experimental results on the Holidays dataset show that the mAP of our algorithm is higher than the 5 mainstream image retrieval algorithms, which efficiently improves the image retrieval accuracy.

Copyright

© 2016, 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  - Hongyan Zhai
PY  - 2016/06
DA  - 2016/06
TI  - Generic Object Regions Matching Based VLAD Model for Image Retrieval
BT  - Proceedings of the 2016 6th International Conference on Machinery, Materials, Environment, Biotechnology and Computer
PB  - Atlantis Press
SP  - 1337
EP  - 1345
SN  - 2352-5401
UR  - https://doi.org/10.2991/mmebc-16.2016.273
DO  - 10.2991/mmebc-16.2016.273
ID  - Zhai2016/06
ER  -