CONTENT GUIDED ATTENTION FOR UNDERWATER IMAGE ENHANCEMENT

Abstract

This thesis addresses the challenge of enhancing underwater image clarity using deep learning techniques, a crucial advancement for applications in marine biology, underwater archaeology, and environmental monitoring. Traditional enhancement methods typically fail to address the harsh light distortions encountered underwater, such as discoloration and blur due to light absorption and scattering This study uses Enhanced- . Encoder FUnIE-GAN (EEF-GAN), which is an updated version of Fast Underwater Image Enhancement comes GAN (FUnIEGAN), which is designed to overcome these challenges by adding new encoder structures The modified encoder uses traditional convolution side convolution difference serves to enhance feature extraction, thereby significantly improving image recognition Empirical results from extensive testing on UIEB dataset show that EEF -GN: peak signal The model outperforms existing models is available in several metrics including -to-noise ratio (PSNR) andstructural similarity index (SSIM) giving aPSNR of22.94 dBandaSSIM of 0.8926, aclear andaccurate underwater image for comparison ato baseline models like WaterNet and UGAN Underscoring its effectiveness to create These findings not only demonstrate the feasibility of using generative anti-nets for real-time image enhancement in complex underwater environments but also demonstrate the potential of such technologies this has in other easily identifiable imaging applications Preferences involve enhancing images and optimizing the model This function by extending the capabilities of deep learning models Contributes to the environment extensive environmental mapping projects, providing new tools for research and conservation efforts.