A Model for Underwater Image Restoration Based on Color Information

With the acceleration of climate change and the proliferation of oceanic exploration technology, there is an increasing need for a comprehensive comprehension of the underwater ecosystem. However, underwater images directly obtained are frequently marred by colour distortion, blurring, low image quality, and inability to reflect the actual scene accurately due to the absorption and scattering of light by aquatic bodies. To address common issues of colour deviation and blurring in underwater images, we suggest a method to restore these images utilising information solely from the captured image. The presented model conducts colour restoration and detail stretching on the input image post-preprocessing, and subsequently delivers the resulting image via a weight-assisted image fusion process. We conducted tests on over 11,270 images from EUVP, R165 and UIEBD datasets. The combined outcomes of the PSNR, SSIM, MSE and UCIQE metrics produced results of 20.9711, 0.8254, 74.3698 and 0.5643, respectively. It is worth noting that certain metrics outperformed comparable models, such as DHE, IBLA, and WB.