CN116523793A - Image map defogging method combining dark channel and convolutional neural network - Google Patents
Image map defogging method combining dark channel and convolutional neural network Download PDFInfo
- Publication number
- CN116523793A CN116523793A CN202310523305.XA CN202310523305A CN116523793A CN 116523793 A CN116523793 A CN 116523793A CN 202310523305 A CN202310523305 A CN 202310523305A CN 116523793 A CN116523793 A CN 116523793A
- Authority
- CN
- China
- Prior art keywords
- image
- map
- calculating
- convolutional neural
- dark channel
- 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
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 21
- 238000005286 illumination Methods 0.000 claims abstract description 35
- 238000002834 transmittance Methods 0.000 claims abstract description 13
- 238000001914 filtration Methods 0.000 claims abstract description 8
- 238000012937 correction Methods 0.000 claims description 7
- 238000003709 image segmentation Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 239000003086 colorant Substances 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 3
- 238000000638 solvent extraction Methods 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 6
- 230000015556 catabolic process Effects 0.000 description 4
- 238000006731 degradation reaction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003139 buffering effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000005577 local transmission Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
Abstract
S1, reading pixel distribution of a map image, calculating a gray value, generating a gray image S2, and calculating a dark channel image of the gray image in an interpolation filtering mode; s3, calculating the estimated transmittance t (x) of the light rays in the image in a priori mode; s4, calculating the atmospheric illumination A of the map image by using a convolutional neural network model; s5, calculating a haze-free image through the atmospheric illumination A and the estimated transmittance t (x) of the light rays, and achieving defogging of the image. The model based on the dark channel can process a large amount of image data in a short time, and is suitable for real-time defogging application; convolutional neural networks can increase atmospheric illumination partitioning details by adding more layers and nodes, and can also be applied to other image processing tasks.
Description
Technical Field
The invention relates to the technical field of geographic information development, in particular to an image map defogging method combining a dark channel and a convolutional neural network.
Background
With the continuous development of digital image technology, image defogging technology has become one of the research hotspots in the field of computer vision. The main purpose of the image defogging technology is to remove haze in the image, so that the image is clearer and more real. The traditional image defogging method is mainly based on a physical model, and the physical characteristics of haze need to be modeled and estimated. However, the effect of this approach is often not satisfactory due to the very complex physical properties of haze. Therefore, in recent years, image defogging techniques based on image prior knowledge are becoming a hot spot of research. The dark channel priori technology is an image defogging algorithm based on image priori knowledge, and utilizes the priori information in a natural image to realize the removal of haze in the image by estimating global atmospheric illumination and local transmission rate in the image. The principle of the dark channel prior technology is simple and effective, and the dark channel prior technology is widely applied to the fields of image defogging, image enhancement, image restoration and the like. Such defogging methods are based on physical model rules, but do not work well when dealing with complex scenes.
Disclosure of Invention
The invention aims to provide an image map defogging method combining a dark channel and a convolutional neural network.
The aim of the invention can be achieved by the following technical scheme:
an image map defogging method combining a dark channel and a convolutional neural network comprises the following steps:
s1, reading pixel distribution of a map image, calculating gray values and generating a gray image;
s2, calculating a dark channel image of the gray level image in an interpolation filtering mode;
s3, calculating the estimated transmittance t (x) of the light rays in the image in a priori mode;
s4, calculating the atmospheric illumination A of the map image by using a convolutional neural network model;
s5, calculating a haze-free image through the atmospheric illumination A and the estimated transmittance t (x) of the light rays, and achieving defogging of the image.
As a further scheme of the invention: in step S1, the gray value is calculated by using a Gamma correction algorithm, where the Gamma correction algorithm formula is as follows:
wherein Gray is a Gray value, and R, G and B are respectively red, green and blue channel colors in the image.
As a further scheme of the invention: in step S3, the calculation formula of the estimated transmittance t≡x is as follows:
wherein x is the image pixel position, C is the color three channels, Ω (x) is the local area, I is the map image, and A is the atmospheric illumination.
As a further scheme of the invention: the process of step S4 includes: dividing the map image;
setting a network model, dividing the network model into two parts of coding and decoding, embedding, dividing and dividing the image information through a convolution kernel, pooling compressed image information, decoding the image information through up-sampling, and adding the decoded and coded image characteristics for splicing;
setting a loss function with boundary weight;
E＝∑ x∈Ω ω(x)log(p ι(x) (x))，
wherein p is ι(x) (x) Loss of function for softmaxThe number, wherein iota E {1, …, K } is the label value of the pixel point, and ω is the weight of the pixel point;
selecting a data set for training a network model, and dividing an atmospheric illumination distribution range through training shadow textures;
and estimating the atmospheric illumination range by using the trained network model, carrying out mean filtering on different atmospheric illumination ranges to obtain the maximum gray value in the different atmospheric illumination ranges, calculating the average value with the maximum gray value in the map image to obtain the atmospheric illumination parameters in the range, and splicing the divided atmospheric illumination A.
As a further scheme of the invention: the calculation formula of the weight omega of the pixel point is as follows:
wherein omega c Is the weight of the balance class proportion, d 1 Is the distance d from the pixel point to the nearest image segmentation area 2 Is the distance, ω, from the pixel to the image segmentation limit that is the second closest thereto 0 =10, σ≡5 pixels.
As a further scheme of the invention: the process of dividing the map image is as follows: the image is first divided according to a size of less than 572 x 572 and then the divided image is filled.
As a further scheme of the invention: in step S5, the formula for calculating the haze-free image is:
the invention has at least one of the following advantages:
(1) High efficiency: the dark channel-based model can process a large amount of image data in a short time, and is suitable for real-time defogging application.
(2) Scalability: convolutional neural networks can increase atmospheric illumination partitioning details by adding more layers and nodes, and can also be applied to other image processing tasks.
(3) The applicability is strong: the model can process haze images of different types and different degrees, and has certain robustness to noise and deformation in the images.
(4) Interpretability: the defogging process of the physical model can be explained and understood through a visual method, which is helpful for in-depth understanding of the principle and method of image defogging.
Drawings
The invention is further described below with reference to the accompanying drawings.
FIG. 1 is a flow chart of the defogging method for an image map according to the present invention;
FIG. 2 is a schematic diagram of the defogging method for the image map of the present invention;
FIG. 3 is a schematic illustration of image segmentation of the present invention;
fig. 4 is a schematic structural diagram of the U-net in the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1-4, the present invention is an image map defogging method combining a dark channel and a convolutional neural network. The method is basically based on an atmospheric degradation model and carries out a responsive defogging treatment. Defogging effect based on the atmospheric degradation model is generally better than defogging algorithm based on image enhancement. In the field of computer vision, a foggy-day image degradation model is generally used to describe the effect of severe weather conditions such as haze on an image, and the model is first proposed by McCartney. The model comprises an attenuation model and an ambient light model. The model expression is:
I(x)＝J(x)t(x)+A(1-t(x))
wherein I (x) is the existing image (to be defogged), x is the pixel position of the image, J (x) is the original defogging image to be recovered, A is atmospheric illumination, t (x) is transmissivity, and is the gray level of the image after transmitting light, and the effect of the image after being influenced by atmospheric light.
Specifically, the image map defogging method combining the dark channel and the convolutional neural network comprises the following steps:
s1, reading the pixel distribution of the map image, calculating the gray value and generating a gray image.
Specifically, in step S1, the RGB color values cannot be simply added directly, but must be converted to physical light power by the power of 2.2. Since RGB values are not simply linear with power, but are power functions, the exponent of this function is called Gamma value, typically 2.2, and this scaling process is called Gamma correction. The gray value is calculated by using a Gamma correction algorithm, and the Gamma correction algorithm formula is as follows:
wherein Gray is a Gray value, and R, G and B are respectively red, green and blue channel colors in the image.
S2, calculating a dark channel image of the gray level image in an interpolation filtering mode.
S3, calculating the estimated transmittance t (x) of the light rays in the image in a priori mode.
Specifically, the transmittance t of air is estimated by the calculated atmospheric illumination a and the original image I, and the transmittance t (x) is represented by a foggy-day image degradation model, wherein the model expression can be changed to:
c is the color three channel, namely C epsilon { r, g, b }, assuming the transmittance of the window is t (x), and is calculated by a local minimum filtering mode:
the minimum value of the color channel is calculated, and then the gray-scale minimum value of the local area omega (x) is calculated on the basis of the obtained minimum value of the channel. The dark channel prior image is calculated as follows:
J dark (x) For the dark channel image without fog state, according to practical calculation summary, the value of the dark channel of the image without fog state is very low, tends to be 0, and as the dark channel image prior, the following exists:
substituting into the formula of t (x) to obtain the predicted value t (x) of the transmissivity:
where Ω (x) is a local area, and I is a map image.
S4, calculating the atmospheric illumination A of the map image by using the convolutional neural network model.
The process of step S4 includes: dividing the map image; referring to fig. 3, the image is divided according to a size smaller than 572×572, the divided image is filled, and the dark channel processed image is divided by filling a blank, so that the size is uniform when the image is input. Filling the blank can be used as excessive buffering between the segmented images, so that the split feeling of the spliced images is reduced.
Setting a network model, dividing the network model into coding and decoding parts, combining with fig. 4, selecting a U-net neural network to extract an atmospheric illumination area, firstly embedding, dividing and dividing image information through a convolution kernel, pooling compressed image information, decoding the image information through up-sampling, and adding decoded and coded image characteristics for splicing;
setting a loss function with boundary weight;
wherein p is ι(x) (x) Is a softmax loss function, wherein iota epsilon {1, …, K } is a label value of a pixel point, and omega is a weight of the pixel point;
selecting a data set for training a network model, and dividing an atmospheric illumination distribution range through training shadow textures; the present invention uses a dataset Shadow Detection/Texture Segmentation Computer Vision Dataset, consisting essentially of texture and Shadow images, which is focused on texture analysis, so that each image sequence contains shadows moving in front of many different texture surfaces.
Dividing the atmospheric illumination range by using the trained network model, carrying out average filtering on different atmospheric illumination ranges to obtain the maximum gray value in the different atmospheric illumination ranges, calculating the average value with the maximum gray value in the map image to obtain the atmospheric illumination parameters in the range, and splicing the divided atmospheric illumination A.
The formula of the weight ω of the pixel point is as follows:
wherein omega c Is the weight of the balance class proportion, d 1 Is the distance d from the pixel point to the nearest image segmentation area 2 Is the distance, ω, from the pixel to the image segmentation limit that is the second closest thereto 0 =10, σ≡5 pixels.
S5, calculating a haze-free image through the atmospheric illumination A and the estimated transmittance t (x) of the light rays, and achieving defogging of the image.
Specifically, in step S5, the restoration formula for calculating the haze-free image is:
the invention has the following advantages:
(1) High efficiency: the dark channel-based model can process a large amount of image data in a short time, and is suitable for real-time defogging application.
(2) Scalability: convolutional neural networks can increase atmospheric illumination partitioning details by adding more layers and nodes, and can also be applied to other image processing tasks.
(3) The applicability is strong: the model can process haze images of different types and different degrees, and has certain robustness to noise and deformation in the images.
(4) Interpretability: the defogging process of the physical model can be explained and understood through a visual method, which is helpful for in-depth understanding of the principle and method of image defogging.
The foregoing describes one embodiment of the present invention in detail, but the description is only a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. All equivalent changes and modifications within the scope of the present invention are intended to be covered by the present invention.
Claims (7)
1. An image map defogging method combining a dark channel and a convolutional neural network is characterized by comprising the following steps:
s1, reading pixel distribution of a map image, calculating gray values and generating a gray image;
s2, calculating a dark channel image of the gray level image in an interpolation filtering mode;
s3, calculating the estimated transmittance t (x) of the light rays in the image in a priori mode;
s4, calculating the atmospheric illumination A of the map image by using a convolutional neural network model;
s5, calculating a haze-free image through the atmospheric illumination A and the estimated transmittance t (x) of the light rays, and achieving defogging of the image.
2. The image map defogging method combining a dark channel and a convolutional neural network according to claim 1, wherein in step S1, the gray value is calculated by using a Gamma correction algorithm, and the Gamma correction algorithm formula is as follows:
wherein Gray is a Gray value, and R, G and B are respectively red, green and blue channel colors in the image.
3. The image map defogging method combining a dark channel and a convolutional neural network according to claim 1, wherein in step S3, the calculation formula of the estimated transmittance t++x is:
wherein x is the image pixel position, C is the color three channels, Ω (x) is the local area, I is the map image, and A is the atmospheric illumination.
4. The image map defogging method combining a dark channel and a convolutional neural network according to claim 1, wherein the process of step S4 comprises: dividing the map image;
setting a network model, dividing the network model into two parts of coding and decoding, embedding, dividing and dividing the image information through a convolution kernel, pooling compressed image information, decoding the image information through up-sampling, and adding the decoded and coded image characteristics for splicing;
setting a loss function with boundary weight;
E＝∑ x∈Ω ω(x)log(p ι(x) (x))，
wherein p is ι(x) (x) Is a softmax loss function, wherein iota epsilon {1, …, K } is a label value of a pixel point, and omega is a weight of the pixel point;
selecting a data set for training a network model, and dividing an atmospheric illumination distribution range through training shadow textures;
and estimating the atmospheric illumination range by using the trained network model, carrying out mean filtering on different atmospheric illumination ranges to obtain the maximum gray value in the different atmospheric illumination ranges, calculating the average value with the maximum gray value in the map image to obtain the atmospheric illumination parameters in the range, and splicing the divided atmospheric illumination A.
5. The image map defogging method combining a dark channel and a convolutional neural network according to claim 4, wherein the weight ω of the pixel points is calculated as follows:
wherein omega c Is the weight of the balance class proportion, d 1 Is the distance d from the pixel point to the nearest image segmentation area 2 Is the distance, ω, from the pixel to the image segmentation limit that is the second closest thereto 0 =10, σ≡5 pixels.
6. The image map defogging method combining a dark channel and a convolutional neural network according to claim 4, wherein the process of dividing the map image is: the image is first divided according to a size of less than 572 x 572 and then the divided image is filled.
7. The image map defogging method combining a dark channel and a convolutional neural network according to claim 1, wherein in step S5, the formula for calculating the defogging image is:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310523305.XA CN116523793A (en) | 2023-05-10 | 2023-05-10 | Image map defogging method combining dark channel and convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310523305.XA CN116523793A (en) | 2023-05-10 | 2023-05-10 | Image map defogging method combining dark channel and convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116523793A true CN116523793A (en) | 2023-08-01 |
Family
ID=87402846
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310523305.XA Pending CN116523793A (en) | 2023-05-10 | 2023-05-10 | Image map defogging method combining dark channel and convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116523793A (en) |
-
2023
- 2023-05-10 CN CN202310523305.XA patent/CN116523793A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108876743B (en) | Image rapid defogging method, system, terminal and storage medium | |
| CN106157267B (en) | Image defogging transmissivity optimization method based on dark channel prior | |
| CN106910175B (en) | Single image defogging algorithm based on deep learning | |
| CN112288658A (en) | Underwater image enhancement method based on multi-residual joint learning | |
| CN107886552B (en) | Mapping processing method and device | |
| CN110288550B (en) | Single-image defogging method for generating countermeasure network based on priori knowledge guiding condition | |
| CN103778900B (en) | A kind of image processing method and system | |
| CN105913390B (en) | A kind of image defogging method and system | |
| CN110544213A (en) | Image defogging method based on global and local feature fusion | |
| CN111709888B (en) | Aerial image defogging method based on improved generation countermeasure network | |
| WO2024060576A1 (en) | Image dehazing method based on dark channel prior | |
| CN105023246B (en) | A kind of image enchancing method based on contrast and structural similarity | |
| CN113034413A (en) | Low-illumination image enhancement method based on multi-scale fusion residual error codec | |
| CN110298809B (en) | Image defogging method and device | |
| CN112164010A (en) | Multi-scale fusion convolution neural network image defogging method | |
| CN110766640B (en) | Image defogging method based on depth semantic segmentation | |
| CN111192213A (en) | Image defogging adaptive parameter calculation method, image defogging method and system | |
| CN113947538A (en) | Multi-scale efficient convolution self-attention single image rain removing method | |
| CN111161159B (en) | Image defogging method and device based on combination of priori knowledge and deep learning | |
| Ling et al. | Quality assessment for synthesized view based on variable-length context tree | |
| CN110349113B (en) | Adaptive image defogging method based on dark primary color priori improvement | |
| CN116523793A (en) | Image map defogging method combining dark channel and convolutional neural network | |
| Liang et al. | Multi-scale and multi-patch transformer for sandstorm image enhancement | |
| CN107203979B (en) | Low-illumination image enhancement method | |
| CN108564535A (en) | A kind of image defogging method based on deep learning |
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 |