Source-linked AI summary
Contrast Enhancement of Brightness-Distorted Images by Improved Adaptive Gamma Correction
Gang Cao, Lihui Huang, Huawei Tian, Xianglin Huang, Yongbin Wang, Ruicong Zhi
TL;DR
AGC works well on many dimmed images but has limitations for globally bright images and dimmed images with local bright regions. This paper combines negative-image processing with CDF-truncated adaptive gamma correction, and reports better or comparable enhancement effects than previous techniques.
Problem
Existing AGC methods do not directly enhance globally bright images and may lose structure in local bright regions of dimmed images.
Source: §1 · p. 3Method
The method applies AGCWD to negative versions of bright images and uses CDF truncation to reconstruct intensity-sensitive adaptive gamma for dimmed images.
Source: §5 · p. 18Results
The proposed scheme achieves better or comparative enhancement effects than previous techniques.
Source: §5 · p. 18Takeaways & Limitations
The method provides contrast enhancement for both bright and dimmed images using the corresponding negative-image and truncated-CDF strategies.
Source: §5 · p. 18AGC remains constrained on globally bright images and dimmed images with local bright regions, the cases the improved method addresses.
Source: §1 · p. 3, §3.2 · p. 7
Abstract
from arXiv · showhide
As an efficient image contrast enhancement (CE) tool, adaptive gamma correction (AGC) was previously proposed by relating gamma parameter with cumulative distribution function (CDF) of the pixel gray levels within an image. ACG deals well with most dimmed images, but fails for globally bright images and the dimmed images with local bright regions. Such two categories of brightness-distorted images are universal in real scenarios, such as improper exposure and white object regions. In order to attenuate such deficiencies, here we propose an improved AGC algorithm. The novel strategy of negative images is used to realize CE of the bright images, and the gamma correction modulated by truncated CDF is employed to enhance the dimmed ones. As such, local over-enhancement and structure distortion can be alleviated. Both qualitative and quantitative experimental results show that our proposed method yields consistently good CE results.
1. Introduction
Brightness-distorted images can impair visual quality and downstream use, while existing contrast-enhancement methods have trade-offs. The paper targets AGC’s failures on globally bright images and dimmed images with local bright regions using negative images and CDF truncation.
1. Introduction
- Contrast enhancement matters because low contrast degrades visual quality and can hinder image analysis, understanding, object recognition, and digital printing.Low or high scene illuminance can reduce contrast quality.Source: §1 · p. 1
- Histogram equalization maps intensities by equalizing the input gray-level CDF, but high histogram peaks can cause over-enhancement.Related approaches include local and brightness-preserving bi-histogram equalization.Source: §1 · p. 2
- Prior CE work includes histogram modification, whose enhancement results are rather sensitive to parameter setting, and a Gaussian mixture model is also proposed; SECE can yield visually improved, pleasing outputs without attractive distortions.Source: §1 · p. 2
- AGCWD relates gamma to the image CDF and performs well on most dimmed images, but can distort bright regions or fail on globally bright images.Low gamma values for large pixel intensities may distort bright regions.Source: §1 · p. 2
- The proposed AGC combines negative images and CDF truncation to address globally bright images and dimmed images with local bright regions.The authors report that experiments verify its effectiveness on both dimmed and bright contrast-distorted images.Source: §1 · p. 3
2. Prior Works on Adaptive Gamma Correction
AGC maps each pixel using a gamma parameter derived from the image’s gray-level CDF. AGCWD weights the histogram to adjust that CDF, but prior AGC methods can distort local bright regions and do not enhance globally bright images.
2. Prior Works on Adaptive Gamma Correction
- AGC derives an adaptive gamma parameter from the gray-level CDF, with gamma decreasing from 1 to 0 as intensity rises.The CDF is formed from the normalized gray-level histogram; the described example uses 8-bit grayscale images with maximum intensity 255.Source: §2 · p. 3
- AGCWD weights the intensity histogram, normalizes the weighted distribution, and derives a corresponding CDF that can be adjusted through α.For α < 1, the weighted histogram becomes relatively flat and smooth at low intensities.Source: §2 · p. 4
- Prior AGC methods can distort bright regions when gamma is too low for large intensities and fail to enhance globally bright images.The globally bright images have a distinctly different pixel-intensity distribution.Source: §2 · p. 4
3. Proposed Method
The method distinguishes bright and dimmed inputs, then applies negative-image AGC to bright images and CDF-truncated AGC to dimmed images.
3.2. AGC via Truncated CDF
- For 8-bit images, τ is set at about 128, approximately half the maximum pixel intensity.The threshold is selected with a trade-off between enhancement quality and technical applicability.Source: §3 · p. 5
3. Proposed Method
- Normal-illuminance images are not targeted; the brightness threshold separates dimmed inputs below it from bright inputs above it.The threshold is set experimentally to balance enhancement quality and technical applicability.Source: §3 · p. 5
3.1. AGC via Negative Image
- For bright images, the method applies AGCWD to a negative image and reverses the enhanced result into positive-image space.Reversing bright pixels creates many low-intensity pixels, allowing the negative image to be treated as dimmed.Source: §3.1 · pp. 5–6
- Direct AGCWD on a bright image yields a worse result because its mapping function is unsuitable for that input.The negative-image approach relies on AGCWD successfully enhancing the dimmed negative.Source: §3.1 · p. 6
- The weighting factor α differs for bright and dimmed images because natural-image histograms may be asymmetric.For the bright-image procedure, α is considered separately.Source: §3.1 · p. 6
3.2. AGC via Truncated CDF
- CDF-truncated AGC floors the adaptive gamma at τ, preventing very low gamma values from over-enhancing bright regions and losing details.The chosen τ is also described as small enough to extend the dynamic range of dimmed pixels.Source: §3.2 · p. 8
3. Proposed Method
- The proposed approaches can also enhance color images by operating on the luminance channel while preserving chrominance in a suitable color space.Source: §3.2 · p. 8
4. Experiments and Discussion
The experiments evaluate the proposed method on bright and dimmed images through visual comparisons and multiple metrics, examining enhancement, distortion, and structure preservation. They also assess its performance on real images and compare its computation time with AGCWD.
4.1. Dataset, Algorithms and Performance Measures
- Performance comparisons test CE algorithms including HE, HM, AGCWD, IMADJ, and SECE using EMEG; simulated inputs apply gamma corrections 0.3 and 2 to 500 test images.The study uses images from the Kodak, BSD500, and UCID standard databases.Source: §4.1 · p. 9, §4.4 · p. 15
- GMSD measures gradient similarity but may miss mean-intensity discrepancies that affect global perceived brightness; PCQI separately measures contrast change, structure distortion, and mean-intensity distortion.Higher Pc indicates greater contrast increase, while higher Ps and Pi indicate less structure and intensity distortion.Source: §4.1 · p. 9
4.2. Qualitative Assessments on Bright Images
- For bright Hats, HE over-enhances shadows and HM/AGCWD provide weak contrast gains, while the proposed method, IMADJ, and SECE boost contrast without annoying artifacts and look consistent with the original.Similar results on Moto, Island, and Tower keep the proposed method visually comparable to IMADJ and SECE; HE and HM do not work well on all samples, and AGCWD performs worst for contrast increase.Source: Figure 6 · p. 10, §4.2 · p. 12
- On bright samples, the method scores higher than AGCWD and HM on EMEG and Pc, with lower GMSD and higher Ps and Pi, and is comparable to IMADJ and SECE.HE achieves high EMEG and Pc but also high GMSD and low Ps and Pi, indicating greater distortions.Source: Table 1 · p. 12, Table 2 · p. 12, §4.2 · p. 12
4.3. Qualitative Assessments on Dimmed Images
- For dimmed Island, the proposed method has Ps=0.992 and Pi=0.943, closer to 1 than HE (Ps=0.989, Pi=0.890) and AGCWD (Ps=0.987, Pi=0.893), although its PCQI is lower due to Pc.Source: §4.3 · p. 14, Table 4 · p. 14
- On dimmed images, HE and AGCWD over-enhance bright regions and lose detail; the proposed method avoids this, while IMADJ leaves images rather black and causes global intensity distortion.The proposed method is comparable with SECE and HM in balancing contrast improvement against distortion; HM also causes global intensity distortion on Hats and Island.Source: Figure 10 · p. 13, §4.3 · pp. 14–15
4. Experiments and Discussion
- On BSD500, the proposed method boosts contrast more than SECE and IMADJ but has slightly worse GMSD and Ps, trading stronger contrast for mildly greater distortion.The paper also describes its performance as comparable with state-of-the-art CE methods.Source: §4.4 · p. 15
- On dimmed images, Pi is 0.902 for the proposed method, versus 0.878 for SECE and 0.834 for IMADJ.Consistent results were also reported on the Kodak and UCID datasets.Source: §4.4 · pp. 15–16
- For bright images, Pc rises from 0.868 (AGCWD) to 1.126 (Prop.), while Ps, Pi rise from 0.976, 0.763 to 0.992, 0.923, respectively.For dimmed images, AGCWD has higher Pc but more serious distortion, reflected in relatively small Ps and Pi values; EMEG and GMSD also support these conclusions.Source: §4.4 · pp. 15–16
- The proposed method takes 28.6 ms per image, comparable to AGCWD and faster than SECE at 38.6 ms; negative-image transformation and CDF truncation do not sharply increase complexity.Table 7 reports average per-image computation times for CE algorithms on BSD500 test images.Source: Table 7 · p. 16, §4.5 · p. 16
4.6. CE on Real Distorted Images
- On real bright images, the proposed method outperforms AGCWD and HE and preserves wrinkle textures in white regions that AGCWD partly removes.The results resemble those of the other approaches overall.Source: §4.6 · pp. 16–17, Figure 11 · p. 17
- On real dimmed images, HE often over-enhances globally and creates local artifacts, while AGCWD performs poorly on white and bright regions and loses detail.The paper gives tower textures and billboard characters as examples of detail better retained by the proposed method.Source: §4.6 · p. 17, Figure 12 · p. 18
- On real dimmed images, IMADJ and SECE tend to under-enhance, whereas the proposed method and HM avoid this defect and generally achieve superior enhancement.The enhanced images from IMADJ and SECE remain rather dim.Source: §4.6 · p. 17, Figure 12 · p. 18
5. Conclusion
The conclusion presents an improved AGC scheme combining negative-image enhancement for bright inputs with CDF-truncated gamma correction for dimmed inputs. Experiments report better or comparable effects without annoying artifacts, while future work targets image types beyond bright and dimmed.
5. Conclusion
- Negative-image processing is used to enhance bright images.This is one component of the improved adaptive gamma correction method.Source: §5 · p. 18
- CDF truncation reconstructs intensity-sensitive adaptive gamma to improve enhancement of dimmed images.The proposed method integrates negative images and CDF truncation.Source: §5 · p. 18
- Extensive qualitative and quantitative experiments show better or comparable enhancement for bright and dimmed inputs without annoying artifacts.The conclusion describes the method as effective and efficient for both input types.Source: §5 · p. 18
- Future work aims to enhance more image types beyond dimmed and bright images.Source: §5 · p. 18