This repository contains code for automatic detection of glasses in near-infrared images implemented by Florian Struck.
This work is licensed under license provided by Hochschule Darmstadt (h_da-License).
Any publications using the code must cite and reference the conference paper [1].
This repository contains 3 different approaches for glasses detection:
- explicit-glasses-identifier - an explicit approach for glasses detection based on edges and reflections
- dl-glasses-identifier - uses deep neuronal network to identify glasses
- statistic-glasses-identifier - uses the BSIF filter and statistical metrics of an image to identify glasses
The repository contains 3 independent projects. Each project has its own structure and dependencies and can therefore be built independently of the other projects. They can be built by running the "make" command in their respective project folders. Afterwards, the executable can be found in PROJECT/build/.
The models used in the paper are available here: (Models)
explicit-glasses-identifier:
- BOOST library (Version >= 1.52)
- OpenCV (Version 2.4)
- glog (Version 0.3.5)
dl-glasses-identifier:
- BOOST library (Version >= 1.52)
- OpenCV (Version 2.4)
- glog (Version 0.3.5)
- Caffe (See http://caffe.berkeleyvision.org/installation.html)
statistic-glasses-identifier:
- BOOST library (Version >= 1.52)
- OpenCV (Version 2.4)
- glog (Version 0.3.5)
- matio (Version 1.5)
Code author: Florian Struck (florian.struck@stud.h-da.de)
- [1] Pawel Drozdowski, Florian Struck, Christian Rathgeb, Christoph Busch: "Detection of Glasses in Near-infrared Ocular Images", in Proc. of the 11th IAPR International Conference on Biometrics (ICB 2018), Queensland, Australia, February 2018.