A Comparative Study of Image Dehazing Techniques for Image Restoration
Total View This Week0
Total View This Week0
Institusion
Universitas Bakrie
Author
Savitri, Khairunnisa
Subject
Image Processing
Datestamp
2023-08-28 04:19:12
Abstract :
Image dehazing is the process of removing haze from images. It is a challenging problem due to the non-linear nature of the haze process and the lack of ground truth data. In recent years, deep learning techniques have been shown to be effective for image dehazing. This research presents a comparative study on image dehazing using CycleGAN and GMAN-Net. CycleGAN is a generative adversarial network that can translate images from one domain to another. GMAN-Net is a generic model-agnostic convolutional neural network that can be used to dehaze images by jointly estimating the transmission map and scene radiance. The study evaluated the performance of CycleGAN and GMAN-Net on a dataset of hazy images with ground truth clear images. The evaluation metrics were PSNR, SSIM, entropy, UIQM, and UCIQE. The results showed that GMAN-Net outperformed CycleGAN on all of the evaluation metrics. GMAN-Net also showed better visual results, with dehazed images that were more accurate and realistic than CycleGAN. The research concludes that GMAN-Net is better for image dehazing than CycleGAN. It is more accurate, realistic, and versatile than CycleGAN.
Image dehazing is the process of removing haze from images. It is a challenging problem due to the non-linear nature of the haze process and the lack of ground truth data. In recent years, deep learning techniques have been shown to be effective for image dehazing. This research presents a comparative study on image dehazing using CycleGAN and GMAN-Net. CycleGAN is a generative adversarial network that can translate images from one domain to another. GMAN-Net is a generic model-agnostic convolutional neural network that can be used to dehaze images by jointly estimating the transmission map and scene radiance. The study evaluated the performance of CycleGAN and GMAN-Net on a dataset of hazy images with ground truth clear images. The evaluation metrics were PSNR, SSIM, entropy, UIQM, and UCIQE. The results showed that GMAN-Net outperformed CycleGAN on all of the evaluation metrics. GMAN-Net also showed better visual results, with dehazed images that were more accurate and realistic than CycleGAN. The research concludes that GMAN-Net is better for image dehazing than CycleGAN. It is more accurate, realistic, and versatile than CycleGAN.
Download
book
BibTex
Latex, Jabref
cloud_download
Original Resource
url resource Institution
