Diffusion in the Dark:
A Diffusion Model for Low-Light Text Recognition

Stanford University
WACV 2024


Abstract

Capturing images is a key part of automation for high-level tasks such as scene text recognition. Low-light conditions pose a challenge for high-level perception stacks, which are often optimized on well-lit, artifact-free images. Reconstruction methods for low-light images can produce well-lit counterparts, but typically at the cost of high-frequency details critical for downstream tasks. We propose Diffusion in the Dark (DiD), a diffusion model for low-light image reconstruction for text recognition. DiD provides qualitatively competitive reconstructions with that of state-of-the-art (SOTA), while preserving high-frequency details even in extremely noisy, dark conditions. We demonstrate that DiD, without any task-specific optimization, can outperform SOTA low-light methods in low-light text recognition on real images, bolstering the potential of diffusion models to solve ill-posed inverse problems.

During training, we randomly select a scale and denoise patches using the low-light measurement and low-resolution estimates of global exposure levels. During inference, we use a multi-scale patch-based approach to gradually reconstruct a full-resolution well-lit image.

Examples of low-light text recognition. DiD consistently recovers high-frequency details that allow for accurate text recognition in extremely noisy and dark conditions.

Ours
LLFlow
Input
LLFlow
Ours
LLFlow
Input
LLFlow
Ours
LLFlow
Input
LLFlow
Ours
LLFlow
Input
LLFlow

Examples of low-light enhancement performance on the LOL dataset. DiD reconstructs high-quality well-lit images from low-light measurements, and preserves high-frequency details better than SOTA can.

BibTeX

@inproceedings{nguyen2024diffusion,
  author    = {Nguyen, Cindy M and Chan, Eric R and Bergman, Alexander W and Wetzstein, Gordon},
  title     = {Diffusion in the Dark: A Diffusion Model for Low-Light Text Recognition},
  journal   = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  year      = {2024},
}