Diving Deep: Comparative Analysis and Design of Neural Network Based Algorithm Selection for Underwater Image Processing

Underwater images captured by remotely operated vehicles (ROVs) often suffer from distortions such as noise, blurring, color degradation, and reduced contrast. These distortions significantly impact image processing techniques essential for tasks like object detection and environmental monitoring. The wide range of available algorithms makes it challenging to select the most suitable one for each specific distorted image, necessitating human judgment to ensure an appropriate choice. This difficulty is further intensified by the need to enhance image quality while operating within tight computational constraints. This article introduces an approach to automatically identify the optimal image-processing algorithm customized to the type of distortion. The approach evaluates the quality of distorted images using both reference and no-reference quality metrics. A supervised classification model built with standard neural networks is employed to automate the algorithm selection process. This approach minimizes the need for human intervention in selecting high-quality output.