Publication Date: 2023/06/03
Abstract: Underwater image enhancement has drawn a lot of attention because of its significance in marine engineering and aquatic robotics. Many techniques for enhancing underwater photographs have been put forth in recent years. Since light propagation underwater and in the atmosphere are different, a particular set of nonlinear visual distortions occur. These distortions are brought on by a variety of factors. Red wavelengths are absorbed in deep water as light travels further, which is why underwater images frequently have a green or blue colour as the dominating hue. Low-contrast, fuzzy, and color-degraded images are the result of such wavelength-dependent attenuation, scattering, and other optical properties of water bodies that cause irregular distortions. The previous CNN-GAN (Generative Adversarial Network) based model for real-time underwater image enhancement is sped up by the upgraded inception model proposed by GAN-Based Underwater Image Enhancement. The suggested model assesses image quality based on its global colour, content, local texture, and style information to construct a perceptual loss function. The dataset being used, called EUVP (Enhancing Underwater Visual Perception), is made up of paired and unpaired collections of underwater images captured by seven distinct cameras under a variety of visibility conditions during maritime explorations and cooperative experiments. The suggested model's accuracy can be increased by learning to enhance and improve underwater image quality from both paired and unpaired training
Keywords: GAN, EUVP, Image enhancement, Underwater images
DOI: https://doi.org/10.5281/zenodo.8001748
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT23MAY1699.pdf
REFERENCES