Modelling saliency attention to predict eye direction by topological structure and earth mover’s distance
View/ Open
Author
Wei, Longsheng
Peng, Jian
Liu, Wei
Wang, Xinmei
Liu, Feng
Note: Order does not necessarily reflect citation order of authors.
Published Version
https://doi.org/10.1371/journal.pone.0181543Metadata
Show full item recordCitation
Wei, Longsheng, Jian Peng, Wei Liu, Xinmei Wang, and Feng Liu. 2017. “Modelling saliency attention to predict eye direction by topological structure and earth mover’s distance.” PLoS ONE 12 (7): e0181543. doi:10.1371/journal.pone.0181543. http://dx.doi.org/10.1371/journal.pone.0181543.Abstract
A saliency attention model for predicting eye direction is proposed in this paper. This work is inspired by the success of the topological structure and Earth Mover’s Distance (EMD) approach. Firstly, we extract visual saliency features such as color contrast, intensity contrast, orientation, and texture. Then, we eliminate disconnected regions in the feature maps to keep topological structure. Secondly, we calculate center surround difference using across-scale EMD between different scales feature maps, rather than utilizing the Difference of Gaussian (DoG), which is used in many other saliency attention models. Thirdly, we across-scale fuse the feature maps in different scale and same feature. Lastly, we take advantage of competition function to calculate feature maps in same feature to form a saliency map, which is use to predict eye direction. Experimental results demonstrated the proposed model outperformed the state-of-the-art schemes in eye direction prediction community.Other Sources
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC5528872/pdf/Terms of Use
This article is made available under the terms and conditions applicable to Other Posted Material, as set forth at http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAACitable link to this page
http://nrs.harvard.edu/urn-3:HUL.InstRepos:34375241
Collections
- HMS Scholarly Articles [17922]
Contact administrator regarding this item (to report mistakes or request changes)