Image Editing based Data Augmentation for Illumination-insensitive Background Subtraction
- Submitting institution
-
University of Northumbria at Newcastle
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 30726655
- Type
- D - Journal article
- DOI
-
10.1108/jeim-02-2020-0042
- Title of journal
- Journal of Enterprise Information Management
- Article number
- -
- First page
- 1
- Volume
- 0
- Issue
- -
- ISSN
- 1741-0398
- Open access status
- Compliant
- Month of publication
- August
- Year of publication
- 2020
- URL
-
-
- Supplementary information
-
https://github.com/dksakkos/illumination_augmentation
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- Yes
- Number of additional authors
-
3
- Research group(s)
-
A - Digital Health and Wellbeing (DH&W)
- Citation count
- 0
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Original contributions include new approaches for training an illumination-invariant deep learning model for background subtraction (BGS). We introduced new approaches to simulating local illumination where a direct light source falls only on a small area of an object from different angles to improve the realism of the lighting effect and to handle global changes in illumination across a whole image. Our approaches generate higher quality segmentation masks that are closer to the ground truth than achieved with other augmentation approaches. To stimulate further research, we shared the implementation of the new framework on GitHub (https://github.com/dksakkos/illumination_augmentation) as an open source project.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -
