Efficient single image-based dehazing technique using convolutional neural networks

• Published in: Multimedia tools and applications 2024-03, Vol.83 (34), p.80727-80749
• Main Authors: Gade, Harish Babu, Odugu, Venkata Krishna, B., Janardhana Rao, B., Satish, N., Venkatram, K., Revathi
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Cameras; Computer Communication Networks; Computer Science; Computer vision; Data Structures and Information Theory; Datasets; Deep learning; Discriminators; Entropy; Fog; Learning; Machine learning; Mean square errors; Multimedia; Multimedia Information Systems; Neural networks; Remote sensing; Special Purpose and Application-Based Systems; Surveillance; Ventilation
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-18784-x

This research proposes a learning-based efficient single-image dehazing method. Dehazing, discriminator, and fine-tuning networks build the end-to-end network model. These three techniques are independently trained on suitable datasets. An end-to-end network architecture improves dehazing. The dehazing network model estimates transmission map, atmospheric light, and parallel convolution layers to analyze the input hazy image. The discrimination network extracted a discriminated dehazing image. Finally, discriminator network model findings are used for fine-tuning. The suggested model is tested using foggy images from various datasets and performance measures including PSNR, SSIM, MSE, and Entropy. The suggested learning-based image dehazing is compared to existing approaches qualitatively and quantitatively. The suggested approach improves PSNR by 34.3% to 3.65% over previous works. The proposed work has a 24.9% higher average SSIM and a 76% lower MSE than current efforts. The entropy of the proposed work is improved by a maximum of 9.38%.