A UIF-METRIC DRIVEN FRAMEWORK FOR MOS PREDICITON AND ADAPTIVE ENHANCEMENT OF UNDERWATER IMAGES

Underwater images often suffer from poor quality due to factors like light scattering and noise. This makesit difficult to see details and colors accurately. This paper presents an adaptive image enhancementframework that dynamically optimizes visual quality using objective metrics and machine learning model.The proposed method evaluates image fidelity through the Underwater Image Fidelity (UIF) score, whichquantifies naturalness, sharpness, and structural similarity between enhanced and reference images. Basedon the UIF score, an adaptive enhancement module selectively applies techniques such as dehazing,denoising, gamma correction, or sharpening to improve image quality. Additionally, visual saliencymodels (VSI and SDSP) are integrated to align enhancements with human perceptual preferences. ARandom Forest Regressor is trained on extracted image features (naturalness, sharpness, and brightness)to predict Mean Opinion Scores (MOS), ensuring quality assessment correlates with subjective humanjudgments. Experimental results demonstrate the framework’s effectiveness in enhancing diverse imagetypes while maintaining perceptual integrity. This work contributes to automated image processing bycombining objective metrics, adaptive enhancement, and machine learning for robust qualityoptimization.