Научная визуализация, 2022, том 14, номер 5, страницы 16 - 44, DOI: 10.26583/sv.14.5.02
SIVDSR-Dhaze: Single Image Dehazing with Very Deep Super Resolution Framework and Its Analysis
Авторы: Sangita Roy1,A, S.S. Chaudhuri2,B
A ECE Department Narula Institute of Technology Kolkata, India
B ETCE Department Jadavpur University Kolkata, India
1 ORCID: 0000-0002-8898-0183, roysangita@gmail.com
2 ORCID: 0000-0001-9849-5766, shelism@rediffmail.com
Аннотация
Adverse climate conditions can affect digital photography and cause issues such as colour shifting, poor visibility, contrast reduction, and fainted appearance due to the scattering of atmospheric Particulate Matter (APM). Estimating an optimum transmission matrix is the key to success for any single image dehazing technique. The use of VDSR 20-weighted Layers ImageNet classifier within l earning based Super Resolution technique x allows improving any image resolution and leads to noise suppression. High Residual Learning gradient clipping ensures fast convergence of the algorithm followed by denoising and artifacts removal as a result of compression. This key introspection has been exercised in improving resolution of the hazy images with an optical image formation model. In addition, we evaluate the benchmark of established images and make results comparisons to the state-of-the-art methods that shows a consistent improvement in accurate scene transmission estimation resulting in clear, natural haze-free radiance. A plausible consistency between execution speed and processing speed has been achieved.
Ключевые слова: VDSR, Dehazing, APM, optical image formation model, SIVDSR, SIVDSR-Dhaze, MOSF, VIA.
