Adaptive Qualityaware underwater image enhancement using multiscale fusion

This project develops an adaptive, quality-aware system for the automated diagnosis of underwater image distortions. The framework utilizes deep learning classification to identify pervasive issues like color casts, haze, blur, and low illumination. It implements and compares three convolutional neural network architectures—a custom CNN, the lightweight MobileNet, and the deep ResNet—trained on the UIEB dataset. Each model learns to classify input images into specific degradation categories with an associated confidence score. This diagnostic step provides a precise quality assessment, establishing a critical foundation for targeted enhancement. The best-performing model is deployed as an interactive web application with a Flask backend and an HTML/CSS/JavaScript frontend, offering users a tool to upload images and receive automatic distortion classification. This work advances underwater image preprocessing by introducing an essential, intelligent quality-assessment layer.

Keywords: Underwater Image Enhancement, Quality-aware Classification, Convolutional Neural Network (CNN), MobileNet, ResNet, UIEB Dataset, Deep Learning, Image Distortion, Flask Application, Web-based Tool.