This content is from the CEWIKI
Low light image enhancement is an important challenge in the development of robust computer vision algorithms. The machine learning approaches to this has been either unsupervised, supervised based on paired dataset or supervised based on unpaired dataset. We analyzed the shortcomings of existing works and they motivated us to answer the following questions.
Does the dependency on a dataset improve/degrade the performance of the learning model.
Can the model be modified to learn from both paired and unpaired dataset.
Can existing classical/learning models be combined based on ablation study over intuition to improve efficacy.
We concentrate our work around improving existing work and incorporating traditional techniques to increase the efficacy of the results. Furthermore, we extend our study to identify whether existing models could be utilized for other related problems such as low-light object classification, controlled light enhancement and low-light video enhancement. Thus our work thrives on existing vision based learning models while extending it's applications to related research problems.
Keywords : Computer vision, Image processing, Image enhancement, Low light, Neural networks.
Link to project page:
- https://teambitecode.github.io/fyp/ (All links and info)
- https://github.com/harshana95/lowlightimageenhance (private at the moment)
- https://github.com/gihanjayatilaka/darkarts (private at the moment)
Harshana Weligampola, Gihan Jayatilaka, Suren Sritharan, Parakrama Ekanayake, Roshan Ragel, Vijitha Herath, Roshan Godaliyadda. An optical physics inspired CNN approach for intrinsic image decomposition.
In 2021 IEEE International Conference on Image Processing (ICIP) (). IEEE.
Harshana Weligampola, Gihan Jayatilaka, Suren Sritharan, Roshan Goldaliyadda, Parakrama Ekanayeka, Roshan Ragel, Vijitha Herath. ,"A Retinex based GAN Pipeline to Utilize Paired and Unpaired Datasets for Enhancing Low Light Images"
Accepted for MERcon 2020
[PDF, PDF (paywalled), Presentation PDF, doi]
Non peer-reviewed documents:
* Code for paper LLCNN * Code for paper Kimmel et al (2003) * Code for paper MSRnet * A set of image processing utilities in python