CN116109509A - Real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction - Google Patents
Real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction Download PDFInfo
- Publication number
- CN116109509A CN116109509A CN202310140633.1A CN202310140633A CN116109509A CN 116109509 A CN116109509 A CN 116109509A CN 202310140633 A CN202310140633 A CN 202310140633A CN 116109509 A CN116109509 A CN 116109509A
- Authority
- CN
- China
- Prior art keywords
- enhancement
- gamma correction
- pixel
- low
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012937 correction Methods 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000005286 illumination Methods 0.000 title claims abstract description 58
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 20
- 230000002708 enhancing effect Effects 0.000 claims abstract description 16
- 238000010586 diagram Methods 0.000 claims abstract description 12
- 230000003044 adaptive effect Effects 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 abstract description 11
- 230000000694 effects Effects 0.000 abstract description 8
- 238000012545 processing Methods 0.000 description 6
- 238000013461 design Methods 0.000 description 5
- 238000013507 mapping Methods 0.000 description 5
- 230000000007 visual effect Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000002411 adverse Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
- G06T5/92—Dynamic range modification of images or parts thereof based on global image properties
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a real-time low-illumination image enhancement method and a system based on pixel-by-pixel gamma correction, wherein the method comprises the following steps: step S1: inputting a low-illumination picture to be enhanced, normalizing the low-illumination picture to be enhanced to be between 0 and 1, and obtaining a normalized low-illumination picture; step S2: inputting the normalized low-illumination picture into a trained lightweight convolutional neural network, and outputting a series of gamma correction parameter graphs consistent with the size of an input image; step S3: based on the gamma correction parameter diagram, performing pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iterative enhancement strategy to obtain a preliminary enhancement result; step S4: and (3) enhancing the contrast by using an adaptive contrast enhancement algorithm on the primary enhancement result, and outputting a final enhancement result of contrast enhancement. The method provided by the invention can be used for executing different enhancement on different areas of the image, has more excellent enhancement effect, and has stronger self-adaptability by combining deep learning.
Description
Technical Field
The invention relates to the field of image enhancement, in particular to a real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction.
Background
A significant factor affecting imaging effects at the time of photographing is the lighting conditions. However, the non-ideal illumination environment is quite common in reality and is further limited by the lack of shooting skills and low quality imaging equipment, under which the captured images often cannot contain clear image content, clear colors and high contrast. The low-light image enhancement task aims to restore these low-light images to a level of quality that is desirable for images taken under normal light conditions. However, efficiently performing this task has challenges such as handling uneven illumination, color bias, and large amounts of noise in imaging, performing fine detail recovery, and achieving high real-time requirements.
The low light enhancement method has very wide application in practice. In addition to improving image quality to meet human aesthetic requirements, it can also assist computer vision related downstream tasks such as night-time autopilot systems to promote their adaptability to low light environments. The existing deep learning-based method is difficult to maintain excellent real-time performance while enhancing high quality. However, high real-time is an indispensable feature in practical deployments because users often cannot tolerate high latency, and computer vision systems often involve the processing of high frame rate video. Therefore, the method for realizing high-quality and high-efficiency enhancement has extremely important significance in practical use.
Traditional low-light image enhancement methods include histogram equalization, gamma correction, and decomposition models based on Retinex theory. Histogram equalization achieves the purpose of improving the contrast of an image by stretching the histogram of the image; gamma correction aims at changing the pixel intensity of an image by a nonlinear function; the Retinex theory states that an image can be decomposed into an illumination component and a reflection component, and an enhancement result can be obtained by optimizing around them. Despite the various improvements made by researchers to these methods, they generally exhibit low generalization, i.e., they cannot robustly process pictures taken in a variety of complex low-light environments. For these pictures, these methods tend to present localized overexposure or underexposure results, or undesirable artifacts, which affect the final visual effect.
In recent years, due to the strong learning ability of convolutional neural networks, a low-light image enhancement method based on a deep learning technology is currently mainstream. These deep learning-based methods can be broadly divided into two categories: a supervised learning method and an unsupervised learning method. The main difference between the two is whether to rely on the supervision information of the "low-light-normal-light" one-to-one paired dataset. Supervised learning methods typically require careful design of the network architecture and then training of the network using low-light-normal light paired image pairs to achieve good enhancement performance. Many supervised learning methods also utilize traditional Retinex theory to guide design methods. The KinD designs three sub-networks that perform Retinex decomposition, remove image degradation, and adjust the illuminance of the image, respectively. Considering that the mapping relation from the low-illumination picture to the normal-illumination picture is one-to-many, the LLFlow introduces a normalization flow to model the complex conditional distribution of the normal-illumination image, thereby relieving the deterministic mapping problem of the reconstruction loss of each pixel. URetinex converts the optimization problem of Retinex decomposition into a learnable problem and simultaneously lightens the image, suppresses noise, and preserves detail over several networks. Although complex models bring about performance improvement, they introduce serious computational burden, i.e., high memory occupation, high computational power and long running time, which affect their practical application. In addition, overfitting training data is also a potential risk to these methods, resulting in their inability to flexibly process diverse low-light images.
In recent years, unsupervised learning methods have been proposed, and these methods only need unpaired data sets or only need low-light pictures to train, so the requirements on the data sets are lower. ZeroDCE learns the parameters of the adjustment curve through a lightweight network structure to adjust the low-light pictures. EnLightenGAN introduces an challenge-generating network model, so unpaired datasets can be used to train the model, increasing the flexibility of the model. RUAS performs augmentation by automatically searching for the optimal network structure by means of neural architecture search to obtain a lightweight model. However, although these methods are lighter than the supervised learning methods, they may produce visual effects of overexposure, underexposure, or color cast when processing complex and diverse low-light pictures, and thus further improvements are needed.
Disclosure of Invention
In order to solve the technical problems, the invention provides a real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction.
The technical scheme of the invention is as follows: a real-time low-light image enhancement method based on pixel-by-pixel gamma correction, comprising:
step S1: inputting a low-illumination picture to be enhanced, normalizing the low-illumination picture to be enhanced to be between 0 and 1, and obtaining a normalized low-illumination picture;
step S2: inputting the normalized low-light picture into a trained lightweight convolutional neural network, and outputting a series of gamma correction parameter graphs consistent with the size of an input image;
step S3: based on the gamma correction parameter diagram, performing pixel-by-pixel gamma correction on the low-illumination input image to be enhanced by using an iterative enhancement strategy to obtain a preliminary enhancement result;
step S4: and enhancing the contrast by using an adaptive contrast enhancement algorithm on the preliminary enhancement result, and outputting a final enhancement result of contrast enhancement.
Compared with the prior art, the invention has the following advantages:
1. the traditional gamma correction applies a single global parameter to all pixels in an image, and the mode can not carry out local adjustment of details, so that overexposure or underexposure results are easy to generate when various realistic low-light pictures are processed, and poor robustness and generalization are shown; meanwhile, the optimal gamma correction parameters of different low lights are also different, and manual setting of the parameters needs to rely on a large amount of experience experiments, so that a large amount of workload is introduced. In order to solve the problems, the invention discloses a real-time low-illumination image enhancement method based on pixel-by-pixel gamma correction, which can perform different enhancement on different areas of an image, so that the enhancement effect is more excellent; meanwhile, by combining deep learning, the gamma correction parameters are learned by a convolutional neural network, so that the method has stronger self-adaptability.
2. Most of the current deep learning methods improve the enhancement performance through complex model design, but also introduce heavy calculation load, and cannot achieve excellent real-time processing performance. The invention combines gamma correction and deep learning, and high quality enhancement can be achieved by an extremely lightweight convolutional neural network. Therefore, the invention occupies less memory resources, requires less calculation amount, has higher speed for processing the low-light pictures, and is more dominant in practical deployment.
Drawings
FIG. 1 is a flow chart of a real-time low-light image enhancement method based on pixel-by-pixel gamma correction in an embodiment of the invention;
FIG. 2 is a schematic diagram of an iterative enhancement process for a pixel-by-pixel gamma corrected low-light image in accordance with an embodiment of the present invention;
FIG. 3a is an original low-light image;
FIG. 3b is a conventional gamma correction;
FIG. 3c is a supervised learning approach;
FIG. 3d is an unsupervised learning method;
FIG. 3e illustrates a real-time low-light enhancement method based on pixel-by-pixel gamma correction mapping according to the present invention;
fig. 4 is a block diagram of a real-time low-light image enhancement system based on pixel-by-pixel gamma correction in accordance with an embodiment of the present invention.
Detailed Description
The invention provides a real-time low-illumination image enhancement method based on pixel-by-pixel gamma correction, which can perform different enhancement on different areas of an image, has more excellent enhancement effect, and has stronger self-adaptability by combining deep learning.
The present invention will be further described in detail below with reference to the accompanying drawings by way of specific embodiments in order to make the objects, technical solutions and advantages of the present invention more apparent.
Example 1
As shown in fig. 1, the method for enhancing a real-time low-illumination image based on pixel-by-pixel gamma correction according to the embodiment of the invention includes the following steps:
step S1: inputting a low-illumination picture to be enhanced, normalizing the low-illumination picture to be enhanced to be between 0 and 1, and obtaining a normalized low-illumination picture;
step S2: inputting the normalized low-illumination picture into a trained lightweight convolutional neural network, and outputting a series of gamma correction parameter graphs consistent with the size of an input image;
step S3: based on the gamma correction parameter diagram, performing pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iterative enhancement strategy to obtain a preliminary enhancement result;
step S4: and (3) enhancing the contrast by using an adaptive contrast enhancement algorithm on the primary enhancement result, and outputting a final enhancement result of contrast enhancement.
In one embodiment, step S1 described above: inputting a low-illumination picture to be enhanced, normalizing the low-illumination picture to be enhanced to be between 0 and 1, and obtaining a normalized low-illumination picture, wherein the method specifically comprises the following steps of:
the low-illumination picture to be enhanced is normalized so that each pixel value thereof is between 0 and 1.
In one embodiment, step S2 above: inputting the normalized low-illumination picture into a trained lightweight convolutional neural network, and outputting a series of gamma correction parameter pictures consistent with the size of an input image, wherein the method specifically comprises the following steps of:
the lightweight convolutional neural network consists of four convolutional layers, wherein the convolutional kernel of each layer is 3 multiplied by 3 and the step length is 1; the number of convolution kernels of the first three layers is 8, and the activation functions are ReLU; the number of convolution kernels of the fourth layer is n, so as to output an n Zhang Gama correction parameter graph G; the activation function is Sigmoid to ensure that the value in the gamma correction parameter diagram falls between 0 and 1; meanwhile, constructing a Loss function training lightweight neural network, wherein the Loss function Loss is a minimized gamma correction parameter graph G to an expected gamma correction parameter graph
L1 distance of (2);
wherein ,
a map of expected gamma correction parameters; (x, y) is a two-dimensional coordinate; dividing the iteration result in the low-illumination picture to be enhanced and the iteration process into a plurality of non-overlapping areas, and solving the average intensity of the areas to obtain an average intensity map M (x, y) of the iteration result in the low-illumination picture to be enhanced and the iteration process; dividing the intensity interval into three parts, i.e. [0, alpha ] 1 )、[α 1 ,α 2] and (α2 ,1]The method comprises the steps of carrying out a first treatment on the surface of the Average intensity falling at [0, alpha 1 ) Image area within the interval, representing the area of the image which is totally black and does not contain any meaningful content, the expected gamma correction parameter map +.>
The gamma parameters corresponding to these regions should be 1 to remain unchanged, preventing unnecessary noise from being brought by meaningless enhancement; average intensity falling at [ alpha ] 1 ，α 2 ]An image area within the section representing a part of the image having content but masked by low light,/->
The corresponding gamma parameter in (2) is beta, so that reasonable enhancement is ensured; the average intensity falls within (. Alpha 2 ，1]An image area within the interval, representing a well-exposed part of the image,/->
The corresponding gamma parameter is 1, so that the poor visual effects such as overexposure and the like of the excessively enhanced belt are prevented without changing the original state;
training the lightweight convolutional neural network through the loss function to obtain a trained lightweight convolutional neural network. Will beThe low-illumination picture to be enhanced is input into a trained lightweight convolutional neural network to obtain n gamma correction parameter pictures (G) consistent with the size of the input image 0 ，G 1 ，...，G n-1 ]。
In one embodiment, the step S3: based on the gamma correction parameter diagram, performing pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iterative enhancement strategy to obtain a preliminary enhancement result, and specifically comprising the following steps:
the formula for pixel-by-pixel gamma correction is as follows:
I en (x，y)＝I(x，y) G(x，y)
wherein ,Ien For enhancing the image, I is the image to be enhanced, and the pixel value range of I is [0,1]G is a gamma correction parameter map, and (x, y) is a two-dimensional coordinate of the image. Based on this, each pixel of the image may be enhanced using a different gamma value.
Gamma correction parameter graph [ G ] based 0 ，G 1 ，...，G n-1 ]Using an iteration enhancement strategy to carry out pixel-by-pixel gamma correction on a low-illumination picture to be enhanced, wherein the formula of the iteration enhancement strategy is as follows:
where i e (0, 1,2,., n } represents an iterative round, X is a low-light input image to be enhanced, Y i As a result of the i-th iteration,
for pixel-by-pixel gamma correction operations, Y n The output primary enhancement result is obtained.
In iteration, the invention is realized by
To amplify the enhancement effect to achieve higher quality enhancement.
As shown in fig. 2, an iterative enhancement flow diagram of a low-light image with pixel-by-pixel gamma correction is shown.
In order to solve the problem of low contrast after enhancement of the low-light picture, the present invention proposes a simple and fast adaptive contrast enhancement method of step S4 as a post-processing procedure to enhance the contrast of the final result.
In one embodiment, step S4 above: the contrast is enhanced by using an adaptive contrast enhancement algorithm to the primary enhancement result, and a final enhancement result of the contrast enhancement is output, which specifically comprises:
the adaptive contrast enhancement formula is as follows:
I out ＝E+(I in -E)×(1+c)
wherein Ω (x, y) is a region centered on (x, y), (s, t) is the two-dimensional coordinates in the region, K is the total number of pixels in the region, E is the result of the mean filtering, I in Is the input image, i.e. the preliminary enhancement result Y n ，I out Is the output image after contrast enhancement, i.e. the final enhancement result, c is the parameter controlling the enhancement amplitude. This step allows the contrast of the image to be enhanced by stretching the dynamic range within the image area.
Fig. 3 a-3 e illustrate a comparison of the real-time low-light enhancement method based on pixel-by-pixel gamma correction mapping proposed by the present invention with other methods. FIG. 3a is an original low-light image; FIG. 3b is a conventional gamma correction; FIG. 3c is a supervised learning approach; FIG. 3d is an unsupervised learning method; fig. 3e illustrates a real-time low-light enhancement method based on pixel-by-pixel gamma correction mapping according to the present invention. The method provided by the invention can be seen to not generate adverse visual effects such as overexposure, underexposure, artificial artifacts, color cast and the like, ensures finer details and structures, and has better enhancement effect. Meanwhile, due to the fact that the extremely light model is used, excellent instantaneity is guaranteed.
The traditional gamma correction applies a single global parameter to all pixels in an image, and the mode can not carry out local adjustment of details, so that overexposure or underexposure results are easy to generate when various realistic low-light pictures are processed, and poor robustness and generalization are shown; meanwhile, the optimal gamma correction parameters of different low lights are also different, and manual setting of the parameters needs to rely on a large amount of experience experiments, so that a large amount of workload is introduced. In order to solve the problems, the invention discloses a real-time low-illumination image enhancement method based on pixel-by-pixel gamma correction, which can perform different enhancement on different areas of an image, so that the enhancement effect is more excellent; meanwhile, by combining deep learning, the gamma correction parameters are learned by a convolutional neural network, so that the method has stronger self-adaptability. Most of the current deep learning methods improve the enhancement performance through complex model design, but also introduce heavy calculation load, and cannot achieve excellent real-time processing performance. The invention combines gamma correction and deep learning, and high quality enhancement can be achieved by an extremely lightweight convolutional neural network. Therefore, the invention occupies less memory resources, requires less calculation amount, has higher speed for processing the low-light pictures, and is more dominant in practical deployment.
Example two
As shown in fig. 4, an embodiment of the present invention provides a real-time low-light image enhancement system based on pixel-by-pixel gamma correction, including the following modules:
the preprocessing module 51 is configured to input a low-illumination picture to be enhanced, normalize the low-illumination picture to be enhanced to a value between 0 and 1, and obtain a normalized low-illumination picture;
the gamma correction parameter map generating module 52 is configured to input the normalized low-illumination picture into a trained lightweight convolutional neural network, and output a series of gamma correction parameter maps with the same size as the input image;
the iteration enhancing module 53 is configured to perform pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iteration enhancing policy based on the gamma correction parameter map, so as to obtain a preliminary enhancing result;
the adaptive contrast enhancement module 54 is configured to enhance the contrast using an adaptive contrast enhancement algorithm on the preliminary enhancement result, and output a final enhancement result of the contrast enhancement.
The above examples are provided for the purpose of describing the present invention only and are not intended to limit the scope of the present invention. The scope of the invention is defined by the appended claims. Various equivalents and modifications that do not depart from the spirit and principles of the invention are intended to be included within the scope of the invention.
The method provided by the invention can be seen to not generate adverse visual effects such as overexposure, underexposure, artificial artifacts, color cast and the like, ensures finer details and structures, and has better enhancement effect. Meanwhile, due to the fact that the extremely light model is used, excellent instantaneity is guaranteed.
Claims (5)
1. A real-time low-light image enhancement method based on pixel-by-pixel gamma correction, comprising:
step S1: inputting a low-illumination picture to be enhanced, normalizing the low-illumination picture to be enhanced to be between 0 and 1, and obtaining a normalized low-illumination picture;
step S2: inputting the normalized low-light picture into a trained lightweight convolutional neural network, and outputting a series of gamma correction parameter graphs consistent with the size of an input image;
step S3: based on the gamma correction parameter diagram, performing pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iterative enhancement strategy to obtain a preliminary enhancement result;
step S4: and enhancing the contrast by using an adaptive contrast enhancement algorithm on the preliminary enhancement result, and outputting a final enhancement result of contrast enhancement.
2. The method for enhancing a real-time low-light image based on pixel-by-pixel gamma correction according to claim 1, wherein said step S2: inputting the normalized low-light picture into a trained lightweight convolutional neural network, and outputting a series of gamma correction parameter pictures consistent with the size of an input image, wherein the gamma correction parameter pictures specifically comprise:
the lightweight convolutional neural network consists of four convolutional layers, wherein the convolutional kernel of each layer of convolutional is 3 multiplied by 3, and the step length is 1; the number of convolution kernels of the first three layers is 8, and the activation functions are ReLU; the number of convolution kernels of the fourth layer is n, so as to output an n Zhang Gama correction parameter graph G; the activation function is Sigmoid to ensure that the value in the gamma correction parameter diagram falls between 0 and 1; meanwhile, constructing a Loss function to train the lightweight convolutional neural network, wherein the Loss function Loss is a value from the minimized gamma correction parameter graph G to the expected gamma correction parameter graph
L1 distance of (2);
wherein ,
a map of expected gamma correction parameters; (x, y) is a two-dimensional coordinate; dividing the iteration result in the low-illumination picture to be enhanced and the iteration process into a plurality of non-overlapping areas, and solving the average intensity of the areas to obtain an average intensity map M (x, y) of the low-illumination picture to be enhanced and the iteration result image; dividing the intensity interval into three parts, i.e. [0, alpha ] 1 )、[α 1 ,α 2] and (α2 ,1]The method comprises the steps of carrying out a first treatment on the surface of the Average intensity falling at [0, alpha 1 ) Image area within the interval, representing the area of the image which is totally black and does not contain any meaningful content, the expected gamma correction parameter map +.>
The gamma parameters corresponding to these regions should be 1; average intensity falling at [ alpha ] 1 ,α 2 ]An image area within the section representing a part of the image having content but masked by low light,/->
The corresponding gamma parameter in (2) is beta, so that reasonable enhancement is ensured; the average intensity falls within (. Alpha 2 ,1]An image area within the interval, representing a well-exposed part of the image,/->
The corresponding gamma parameter is 1;
after the low-illumination picture is input into the trained lightweight convolutional neural network, a series of gamma correction parameter graphs (G) consistent with the size of the input image are obtained 0 ，G 1 ，...，G n-1 ]。
3. The method for enhancing a real-time low-light image based on pixel-by-pixel gamma correction according to claim 1, wherein said step S3: based on the gamma correction parameter diagram, performing pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iterative enhancement strategy to obtain a preliminary enhancement result, which specifically comprises:
the pixel-by-pixel gamma correction formula is as follows:
I en (x，y)＝I(x，y) G(x，y)
wherein ,Ien For enhancing the image, I is the image to be enhanced, and the pixel value range of I is [0,1]G is a gamma correction parameter diagram, and (x, y) is a two-dimensional coordinate of an image;
based on the gamma correction parameter map [ G ] 0 ，G 1 ，...，G n-1 ]And carrying out pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iterative enhancement strategy, wherein the iterative enhancement strategy has the following formula:
wherein i e {0,1,2,..n } represents an iterative round, X is the low-light picture to be enhanced, Y i As a result of the i-th iteration,
for pixel-by-pixel gamma correction operations, Y n The output primary enhancement result is obtained.
4. The method for enhancing a real-time low-light image based on pixel-by-pixel gamma correction according to claim 1, wherein said step S4: and enhancing the contrast by using an adaptive contrast enhancement algorithm to the preliminary enhancement result, and outputting a final enhancement result of contrast enhancement, wherein the method specifically comprises the following steps of:
the adaptive contrast enhancement formula is as follows:
I out ＝E+(I in -E)×(1+c)
wherein Ω (x, y) is a region centered on (x, y), (s, t) is the two-dimensional coordinates in the region, K is the total number of pixels in the region, E is the result of the mean filtering, I in Is the input image, i.e. the preliminary enhancement result Y n ，I out Is the output image after contrast enhancement, i.e. the final enhancement result, c is the parameter controlling the enhancement amplitude.
5. A real-time low-light image enhancement system based on pixel-by-pixel gamma correction, comprising the following modules:
the preprocessing module is used for inputting the low-illumination picture to be enhanced, normalizing the low-illumination picture to be enhanced to be between 0 and 1, and obtaining a normalized low-illumination picture;
the gamma correction parameter map generating module is used for inputting the normalized low-illumination picture into a trained lightweight convolutional neural network and outputting a series of gamma correction parameter maps with the same size as the input image;
the iteration enhancement module is used for carrying out pixel-by-pixel gamma correction on the low-illumination picture to be enhanced by using an iteration enhancement strategy based on the gamma correction parameter map to obtain a preliminary enhancement result;
and the self-adaptive contrast enhancement module is used for enhancing the contrast by using a self-adaptive contrast enhancement algorithm on the primary enhancement result and outputting a final enhancement result of contrast enhancement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310140633.1A CN116109509A (en) | 2023-02-13 | 2023-02-13 | Real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310140633.1A CN116109509A (en) | 2023-02-13 | 2023-02-13 | Real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116109509A true CN116109509A (en) | 2023-05-12 |
Family
ID=86257955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310140633.1A Pending CN116109509A (en) | 2023-02-13 | 2023-02-13 | Real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116109509A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116957990A (en) * | 2023-09-19 | 2023-10-27 | 东莞市华复实业有限公司 | Outdoor water meter data vision acquisition method |
| CN117635478A (en) * | 2024-01-23 | 2024-03-01 | 中国科学技术大学 | Low-light image enhancement method based on spatial channel attention |
-
2023
- 2023-02-13 CN CN202310140633.1A patent/CN116109509A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116957990A (en) * | 2023-09-19 | 2023-10-27 | 东莞市华复实业有限公司 | Outdoor water meter data vision acquisition method |
| CN116957990B (en) * | 2023-09-19 | 2023-12-08 | 东莞市华复实业有限公司 | Outdoor water meter data vision acquisition method |
| CN117635478A (en) * | 2024-01-23 | 2024-03-01 | 中国科学技术大学 | Low-light image enhancement method based on spatial channel attention |
| CN117635478B (en) * | 2024-01-23 | 2024-05-17 | 中国科学技术大学 | Low-light image enhancement method based on spatial channel attention |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116109509A (en) | Real-time low-illumination image enhancement method and system based on pixel-by-pixel gamma correction | |
| CN110232661B (en) | Low-illumination color image enhancement method based on Retinex and convolutional neural network | |
| CN113052814B (en) | Dim light image enhancement method based on Retinex and attention mechanism | |
| CN111915513B (en) | Image denoising method based on improved adaptive neural network | |
| Wang et al. | Joint iterative color correction and dehazing for underwater image enhancement | |
| KR102095443B1 (en) | Method and Apparatus for Enhancing Image using Structural Tensor Based on Deep Learning | |
| Yan et al. | Enhanced network optimized generative adversarial network for image enhancement | |
| CN113658057A (en) | Swin transform low-light-level image enhancement method | |
| Lv et al. | Low-light image enhancement via deep Retinex decomposition and bilateral learning | |
| CN112614063B (en) | Image enhancement and noise self-adaptive removal method for low-illumination environment in building | |
| Xu et al. | Gan based multi-exposure inverse tone mapping | |
| CN115035011A (en) | Low-illumination image enhancement method for self-adaptive RetinexNet under fusion strategy | |
| Chen et al. | CERL: A unified optimization framework for light enhancement with realistic noise | |
| Hua et al. | Low-light image enhancement based on joint generative adversarial network and image quality assessment | |
| CN117422653A (en) | Low-light image enhancement method based on weight sharing and iterative data optimization | |
| He et al. | Low-light image enhancement combined with attention map and u-net network | |
| CN116645296A (en) | Non-uniform low-light image enhancement method and system under zero reference sample | |
| CN111161189A (en) | Single image re-enhancement method based on detail compensation network | |
| Oh et al. | Residual dilated u-net with spatially adaptive normalization for the restoration of under display camera images | |
| CN114663300A (en) | DCE-based low-illumination image enhancement method, system and related equipment | |
| CN106952243A (en) | UUV Layer Near The Sea Surface infrared image self adaptation merger histogram stretches Enhancement Method | |
| Wen et al. | Self-Reference Deep Adaptive Curve Estimation for Low-Light Image Enhancement | |
| Song et al. | NRNet: Retinex Decomposition with Realistic Noise | |
| Gupta et al. | Del-Net: A Single-Stage Network for Mobile Camera ISP | |
| Fan et al. | Self-Supervised Low-Light Image Enhancement based on Retinex Model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |