One-Step Diffusion Model for Dark Burst Image Enhancement


SARIGÜL M., Karacan L., ATMIŞ M.

12th International Conference on Mechatronics and Robotics Engineering, ICMRE 2026, Oldenburg, Almanya, 2 - 04 Mart 2026, ss.284-288, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/icmre69538.2026.11533925
  • Basıldığı Şehir: Oldenburg
  • Basıldığı Ülke: Almanya
  • Sayfa Sayıları: ss.284-288
  • Anahtar Kelimeler: Diffusion Models, Low-Light Enhancement, Multi-frame Fusion, RAW Image Enhancement, Robot vision
  • Çukurova Üniversitesi Adresli: Evet

Özet

Robust visual perception in extreme low-light environments is a fundamental requirement for autonomous robotic systems operating in unstructured or nighttime conditions. However, conventional enhancement methods often introduce significant computational latency, hindering real-time decision-making and control. This paper presents a high-efficiency dark burst image enhancement framework leveraging the speed of one-step diffusion models. By adapting the Stable Diffusion Turbo architecture to process multi-frame sequences, we achieve rapid image restoration without the iterative overhead of standard diffusion processes. Validation on the Sony SID dataset demonstrates that our method provides a state-of-the-art LPIPS score of 0.221 while maintaining structural fidelity with a PSNR of 29.22 and an SSIM of 0.87. The proposed approach significantly improves perceptual clarity and detail preservation, offering a computationally viable solution for enhancing the situational awareness of robots in lowvisibility environments.