Abstract :

Underwater Photographs Usually Look Green Or Blue, Flat And Hazy. Water Absorbs And Scatters Light Differently At Different Wavelengths, And This Limits How Useful The Images Are In Marine Biology, Underwater Archaeology, Pipeline Inspection And Vision For Autonomous Vehicles. In This Paper We Present A Hybrid Enhancement Framework That Is Guided By Physics And Assisted By A CNN, Organised As A 14-stage Pipeline. Classical Restoration Methods Estimate Scene Transmission With The Dark-channel Prior (DCP). We Instead Replace The Transmission/depth Ramp With A Dense Monocular Depth Map From A MiDaS/DPT Network, Which Gives A Physically Consistent Depth Cue That Is Largely Unaffected By The Strong Colour Cast Of Underwater Scenes. The Depth Map Drives A Physics-based Decomposition And Colour-restoration Stage, And A Lightweight Three-layer Convolutional Residual-refinement Network Then Corrects The Artefacts This Stage Leaves Behind. A Second, Independent Branch Handles Illumination Estimation, Depthbased Colour Compensation, Scattering Compensation And Contextual Sharpening. We Merge The Two Branches Using Multi-cue Weight Maps (Laplacian Contrast, Saliency And Saturation) And A Three-level Laplacian Pyramid Fusion, Which Gives An Output With High Contrast And Natural Colour. The Pipeline Is Evaluated With A Five-metric Module Made Up Of Two No-reference Measures (UIQM, UCIQE) And Three Full-reference Measures (PSNR, SSIM, LPIPS). On The Representative Test Image, UIQM Rose From −1.3716 To 2.5098 And UCIQE From 0.4226 To 0.6138. Relative To The Original, The Enhanced Output Reached PSNR = 12.8511 DB, SSIM = 0.8289 And LPIPS = 0.0007. Colourfulness, Sharpness And Contrast Improve Markedly, While The Structure And Perceptual Content Of The Source Image Are Largely Retained.

Published:

28-9-2026

Issue:

Vol. 26 No. 9 (2026)

Page Nos:

459-467

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite