Abstract :

Photographs Taken Underwater Look Wrong In Predictable Ways. Water Absorbs Red Light First, Scatters The Rest, And Leaves A Colour Cast, A Veil Of Haze And Very Little Contrast. In This Paper We Describe A Pipeline That Tries To Undo These Effects From A Single Image, Without Depth Maps, Special Hardware Or Training Data. It Starts By Correcting The Dominant Colour Cast With A Stretch Whose Clipping Depends On How Weak Each Channel Is, Then Applies A Power-law Correction For Scattering, And Builds A Second, Sharpened Version Of The Result With An Unsharp Mask. The Two Versions Are Merged In A Three-level Laplacian Pyramid, Guided By Weight Maps Based On Local Contrast, Saliency And Saturation. The Code Also Has An Evaluation Module That Reports UIQM And UCIQE (no Reference) And PSNR And SSIM (full Reference), In Single-image Or Batch Mode, So That Datasets Such As UIEB Can Be Used. On A Test Image With A Very Strong Green Cast, UIQM Went From −1.3716 To 2.5098 And UCIQE From 0.4226 To 0.6138, An Increase Of 45.2%. The Green Veil Disappears And Details Such As The Beard, The Wetsuit Texture And The Chest Insignia Become Visible Again. We Had No Reference Image For This Run, So The PSNR (12.85 DB) And SSIM (0.8289) Only Tell How Far The Output Moved From The Input; They Say Nothing About Accuracy, And We Explain How The Batch Mode Gives Proper Full-reference Scores.

Published:

14-10-2026

Issue:

Vol. 26 No. 10 (2026)

Page Nos:

48-56

Section:

Articles

License:

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

How to Cite