ARTFEED — Contemporary Art Intelligence

CoRe-UIE: A New Framework for Underwater Image Enhancement

ai-technology · 2026-08-11

A new framework named CoRe-UIE has been developed by researchers to enhance underwater images, tackling the issue of coexisting and region-specific degradations. Underwater photography frequently encounters various problems, including color distortion, scattering haze, texture loss, and inconsistent lighting, which can occur simultaneously and differ across regions. Conventional restoration techniques often fail to accommodate these diverse degradation types. CoRe-UIE introduces a collaborative framework focused on degradation, featuring a content-preserving shared expert alongside four specialized experts for color correction, scattering reduction, texture restoration, and illumination enhancement. These experts, while sharing the same architecture, operate with distinct parameters and are assigned to regions based on input-derived degradation signals and region-adaptive Top-k routing. Additionally, a Hilbert-Schmidt Independence Criterion is integrated to boost performance. The paper can be found on arXiv under the identifier 2608.08965.

Key facts

  • CoRe-UIE is a new framework for underwater image enhancement.
  • It addresses coexisting and region-wise degradations in underwater images.
  • Degradations include color distortion, scattering haze, texture attenuation, and uneven illumination.
  • The framework uses a content-preserving shared expert and four routed experts.
  • Routed experts handle color correction, scattering suppression, texture recovery, and illumination protection.
  • Experts share architecture but have independent parameters.
  • Region-adaptive Top-k routing assigns experts to different regions.
  • The paper is available on arXiv with identifier 2608.08965.

Entities

Institutions

  • arXiv

Sources