Systematic Analysis of Underwater Image Dehazing and Object Detection

Dehazing underwater images is crucial for several underwater applications, including marine biology, exploration, and monitoring. Because of colour distortion, light attenuation, and scattering, underwater imaging presents particular difficulties. This research presents a systematic survey on using transformers and different Generative Adversarial Networks (GANs) for dehazing underwater photos. This survey study looks into transformer-based models, analyzes GAN designs (CycleGAN, Pix2Pix, and UWGAN) in detail, and compares them using a range of metrics, such as Mean Squared Error (MSE), Structural Similarity Index Measure (SSIM), Frechet Inception Distance (FID), and F1-score. Further analysis was carried out using CLAHE on the best-performing models. Our findings emphasize the benefits and drawbacks of the different designs that were employed as well as prospective directions for further study and advancement.