Patch-based Cascaded Underwater Image Enhancement via HSV-Domain Patch Processing and Full-Resolution RGB Refinement with Heterogeneous Skip Connections

dc.contributor.authorGautam, Aditi
dc.date.accessioned2026-08-20T10:58:17Z
dc.date.available2026-08-20T10:58:17Z
dc.date.issued2026
dc.description.abstractUnderwater images often suffer from severe degradation due to light scattering, color attenuation, and non-uniform illumination. This paper proposes a cascaded two-stage deep learning framework that combines localized enhancement in the HSV color space with global refinement in the RGB domain. In Stage-1, overlapping 256 × 256 patches are enhanced independently to correct region-specific color and illumination distortions, followed by cosine-based blending for seamless reconstruction. Stage-2 performs full-resolution refinement using a residual learning-based network with multi-scale feature extraction, deep supervision, and channel attention to enforce global consistency and improve perceptual quality. Experimental results demonstrate that the proposed method achieves superior performance in terms of PSNR, SSIM, UIQM, and UCIQE, producing visually coherent images with improved color fidelity and structural preservation. The framework provides an effective and robust solution for underwater image enhancement across diverse degradation conditions.
dc.identifier.urihttp://nits.ndl.gov.in/handle/123456789/104
dc.language.isoen
dc.publisherNational Institute of Technology, Silchar
dc.titlePatch-based Cascaded Underwater Image Enhancement via HSV-Domain Patch Processing and Full-Resolution RGB Refinement with Heterogeneous Skip Connections
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
ADITI GAUTAM_M.Tech Thesis - Aditi Gautam.pdf
Size:
11.8 MB
Format:
Adobe Portable Document Format
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed upon to submission
Description: