PROJECT TITLE :

Accurate Transmission Estimation for Removing Haze and Noise From a Single Image

ABSTRACT:

Image noise frequently results in depth-dependent artefacts while dehazing a single image. For noisy and foggy inputs, most known dehazing algorithms employ a two-step procedure, which results in erroneous transmission maps and low-quality scene illumination. The transmission map and scene radiance can now be recovered from a single image by using an unique variational model, which we describe as a solution to these issues. Non-local regularisation that is transmission-aware prevents noise amplification while maintaining fine features in the recovered image by adaptively reducing noise. Furthermore, we introduce a semantic-guided regularisation to smooth out the transmission map while maintaining depth inconsistencies at the frontiers of various objects. The segmentation map is also optimised using an alternating approach that incorporates both the transmission map and the scene radiance. It has been shown that the suggested technique outperforms current dehazing algorithms on noisy and hazy photos after extensive testing.