Source-linked AI summary
LIME: A Method for Low-light IMage Enhancement
Xiaojie Guo
TL;DR
Low-light images have low visibility that may degrade vision algorithms designed for high-quality inputs. LIME estimates illumination from RGB maxima and refines the map using structure-aware smoothing; experiments report advantages over several state-of-the-art methods.
Problem
Low-light images have poor visibility that may degrade the performance of algorithms designed for high-visibility inputs.
Source: §1 · p. 1Method
LIME estimates each pixel’s illumination from its maximum R, G, and B intensity and refines the initial map using structure-aware smoothing.
Source: §1 · pp. 1–2Results
Takeaways & Limitations
The authors present improved illumination consistency as a way to improve visual quality and offer contemporary vision applications more reliable inputs.
Source: §4 · p. 7Enhancement can reveal intensive noise hidden in very low-light inputs, so denoising is required.
Source: §3 · p. 6
Abstract
from arXiv · showhide
When one captures images in low-light conditions, the images often suffer from low visibility. This poor quality may significantly degrade the performance of many computer vision and multimedia algorithms that are primarily designed for high-quality inputs. In this paper, we propose a very simple and effective method, named as LIME, to enhance low-light images. More concretely, the illumination of each pixel is first estimated individually by finding the maximum value in R, G and B channels. Further, we refine the initial illumination map by imposing a structure prior on it, as the final illumination map. Having the well-constructed illumination map, the enhancement can be achieved accordingly. Experiments on a number of challenging real-world low-light images are present to reveal the efficacy of our LIME and show its superiority over several state-of-the-arts.
1 Introduction
Low-light images obscure scene details and may hurt algorithms designed for high-visibility inputs. LIME estimates illumination and refines it with structure-aware smoothing.
1 Introduction
- Low-light images can bury scene details and may hurt algorithms designed for high-visibility inputs, motivating image enhancement.The paper illustrates the problem with details, such as wall paintings, nearly buried in darkness.Source: §1 · p. 1
- Direct amplification can saturate bright regions; histogram equalization avoids that issue but risks over- or under-enhancement, while was unattractive in the authors’ tests.The authors report that performed poorly in both efficiency and visual quality in their test.Source: §1 · p. 1
- Prior methods dehazed inverted low-light images, with adding over-segmentation and adaptive denoising; the authors say their underlying model lacks physical explanation.The paper aims to connect that model to a more physically meaningful one.Source: §1 · p. 1
- LIME estimates each pixel’s illumination from the maximum R, G, and B intensity, then refines the map with structure-aware smoothing.The authors report that experiments on challenging images demonstrate advantages over other state-of-the-art methods.Source: §1 · pp. 1–2
2 Proposed Method
LIME models a low-light image as the product of a recovered scene and an illumination map, then refines that map with structure-aware smoothing. It also applies gamma adjustment, denoising, and recomposition to improve the recovered image.
2 Proposed Method
- The method treats the captured image as the element-wise product of the desired recovery and an illumination map, whose estimation is key to recovering the image.The paper contrasts this model with an inverted-image haze model, whose physical meaning it says is less direct.Source: §2 · pp. 2–3
2.1 Initial Illumination Map Estimation
- LIME initially estimates illumination using a simple approach, drawing on the dark-channel prior for the inverted image to address non-uniform illumination.The paper notes that Max-RGB boosts only global illumination and that many prior approaches consider local consistency.Source: §2.1 · p. 3
2.2 Illumination Map Refinement
- Structure-aware smoothing refines the initial illumination map to preserve overall structure while smoothing texture; the paper reports an advantage over other map-refinement methods.The objective balances fidelity to the initial map against weighted smoothness; setting all weights to 1 yields classic ℓ2 total variation, which cannot distinguish strong edges from texture.Source: §2.2 · p. 4, Figure 4 · p. 4
2.3 Other Operations
- After refining the map, LIME uses γ = 0.8 in later experiments; denoising addresses amplified noise, and recomposition adapts processing across regions.The paper uses BM3D and describes recomposition as avoiding dark-region under-denoising and bright-region over-smoothing.Source: Figure 6 · p. 5, §2.3 · p. 6
3 Experimental Results
LIME is compared with several enhancement baselines on low-light images, with reported visual-quality advantages over HE, DeHz, and NPE. Very low-light enhancement also exposes hidden noise, for which the authors apply BM3D denoising.
3 Experimental Results
- Comparisons include HE, AHE, Gamma Correction, DeHz, and NPE, with all codes run in Matlab to support fair time comparisons.The DeHz code was not publicly available when the paper was prepared but was described as easy to implement based on.Source: §3 · p. 6
- LIME provides better visual quality than HE, DeHz, and NPE in the reported comparison, while being comparable to or more efficient than DeHz and much more efficient than NPE.It takes longer than HE, AHE, and GC; AHE and GC are omitted from later comparisons because of problems in the example.Source: §3 · p. 6
- Very low-light inputs reveal hidden noise after LIME enhancement, and the authors show improved visual quality after applying BM3D denoising.The paper describes noise emergence as an issue encountered by almost all existing low-light enhancement algorithms.Source: §3 · p. 6
4 Conclusion
The paper presents LIME as an efficient and effective approach to low-light image enhancement. It relies on structure-aware smoothing to improve illumination consistency, and experiments report advantages over several state-of-the-art alternatives.
4 Conclusion
- LIME is presented as an efficient and effective method for improving low-light image visual quality and offering vision applications reliable inputs.Source: §4 · p. 7
- Structure-aware smoothing is developed to improve illumination consistency, which the paper identifies as central to enhancement quality.Source: §4 · p. 7
- Experiments report advantages for LIME compared with several state-of-the-art alternatives.Source: §4 · p. 7