Research Article
Single Image Dehazing Through Feed Forward Artificial Neural Network
@INPROCEEDINGS{10.1007/978-3-031-28975-0_9, author={K. Soni Sharmila and A. V. S. Asha and P. Archana and K. Ramesh Chandra}, title={Single Image Dehazing Through Feed Forward Artificial Neural Network}, proceedings={Cognitive Computing and Cyber Physical Systems. Third EAI International Conference, IC4S 2022, Virtual Event, November 26-27, 2022, Proceedings}, proceedings_a={IC4S}, year={2023}, month={3}, keywords={Dark channel prior Dehazing Transmission map Multilayer perceptron Restoration}, doi={10.1007/978-3-031-28975-0_9} }- K. Soni Sharmila
A. V. S. Asha
P. Archana
K. Ramesh Chandra
Year: 2023
Single Image Dehazing Through Feed Forward Artificial Neural Network
IC4S
Springer
DOI: 10.1007/978-3-031-28975-0_9
Abstract
Due to light scattering by air particles, images taken under bad lighting conditions (such as haze, fog, mist, or smog) have a reduced level of visibility. One blurry image can be made visible again using single image dehazing techniques. Due to the ill-posedness of the Single image dehazing problem, it is difficult to solve. Even though, the Dark channel prior (DCP) has been the most prevalent method for image dehazing algorithms, but it suffers from huge computational time and picture quality. The accurate transmission map estimation is one of the best preferred ways to achieve the least computational time and high picture quality in DCP. Hence, this work is focused on constructing transmission map of a hazy image based on feed forward artificial neural network. The recommended method uses a feed forward ANN to conduct the transmission map straight from the minimum channel and a normalizing procedure to increase the re-covered image information. With a training set of eighty images, the network is trained by means of mean square error (MSE). The proposed method utilizing peak signal to noise ratio (PSNR) and structural similarity (SSIM) index measures are used to assess the restoration quality. The investigational conclusions have demonstrated that, the suggested method outperforms the dehazing of an input image without degrading visible quality (PSNR = 69.16, SSIM index = 0.8913). In addition, the proposed method is suited for real-time applications given the average computing time it achieves is 1.03 s.
