Single Image Very Deep Super Resolution (SIVDSR) Dehaze

Abstract

Adverse climate conditions affect digital photography causing colour shifting, poor visibility, contrast reduction, and fainted appearance due to the scattering of atmospheric Particulate Matter (APM). To get an optimum transmission matrix is the key success of any single image dehazing technique. Deep Learning based Super Resolution technique with VDSR 20-weighted Layers ImageNet classifier improves any image resolution leading to noise suppression. High Residual Learning gradient clipping makes the algorithm converge fast with denoising and removal of artifacts by compression. This key observation has been exercised in improving resolution of the hazy images with an optical image formation model. In addition, benchmark established images are evaluated and their comparisons to the state-of-the-art methods show a consistent improvement in accurate scene transmission estimation resulting in clear, natural haze-free radiance. A good balance between execution speed and processing speed has been achieved.