Single Underwater Image Restoration Using Variational Framework Guided by Imaging Model with Noise

This project presents an effective method for underwater image restoration by addressing the common challenges of distortion and noise in underwater environments. The proposed approach includes three main stages: background light estimation, transmission map calculation, and image restoration using a variational framework. Initially, background light is estimated by identifying the brightest pixels from the dark channel, which represent the scene's background illumination. This step is crucial for enhancing the contrast of submerged images. Following this, a transmission map is calculated using a simplified dark channel prior to model the haze effect in underwater scenes. To refine the initial transmission map, a first-order guided filter is applied, resulting in a more accurate representation of image depth and object visibility. Noise simulation is introduced to mimic real-world underwater noise, with noise removal performed using median filtering for improved clarity. Finally, the restored image is generated using a variational framework based on the transmission map and background light. The result is a clear, visually enhanced image that mitigates haze and noise while improving feature visibility. This method can be particularly useful for underwater imaging applications such as marine biology, underwater exploration, and underwater robotics. The implementation is carried out in MATLAB, leveraging its image processing capabilities.

Keywords: Underwater Images Dataset, Pre-Processing, Image Processing Techniques and Enhancement.