CN109978798A - Ghost image based on image gradient sparsity reflects minimizing technology - Google Patents
Ghost image based on image gradient sparsity reflects minimizing technology Download PDFInfo
- Publication number
- CN109978798A CN109978798A CN201910276482.6A CN201910276482A CN109978798A CN 109978798 A CN109978798 A CN 109978798A CN 201910276482 A CN201910276482 A CN 201910276482A CN 109978798 A CN109978798 A CN 109978798A
- Authority
- CN
- China
- Prior art keywords
- image
- reflection
- ghost image
- model
- ghost
- 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.)
- Granted
Links
- 238000005516 engineering process Methods 0.000 title claims abstract description 14
- 238000000034 method Methods 0.000 claims abstract description 35
- 239000011159 matrix material Substances 0.000 claims description 20
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 239000010410 layer Substances 0.000 description 4
- 239000011521 glass Substances 0.000 description 3
- 238000000926 separation method Methods 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000002620 method output Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000002356 single layer Substances 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of, and the ghost image based on image gradient sparsity reflects minimizing technology, including building ghost image reflection model；Ghost image reflection model is converted into ghost image reflection Variation Model；Ghost image reflection Variation Model is solved；Model result is exported in a manner of picture, the image after obtaining final ghost image reflection removal.The method of the present invention can largely shorten the processing time of algorithm, retain the structure of image, the detailed information such as texture, to obtain better comprehensive performance as far as possible while removing image reflection.
Description
Technical field
Present invention relates particularly to a kind of, and the ghost image based on image gradient sparsity reflects minimizing technology.
Background technique
Return ghost image is passed by entering image by glass-reflected in the light of the object of the same side with imaging sensor
Shooting showcase is common in caused by sensor, in life, indoors the scene outdoor through glass shooting with camera.From object
It is seen in angle of science, when light enters another medium from a kind of medium, a part of light will necessarily reflect.How
Using mathematical physics knowledge by a part of reflection light of this in return ghost image be depicted come, and by reflected image and transmission
Image separation, becomes computer vision field urgent problem to be solved.
Image restoration is the basic project of image procossing and computer vision, this problem is unfolded in numerous studies personnel
Further investigation.In the reflected image of shooting, common reflection type: single layer reflection, bilayer reflection and reflection multilayer.For going
Except the method for reflection there are more image inputs to go to reflect, goes to reflect based on single image input, be based on utilizing its part under simple scenario
Feature carries out reflection separation.The above method establishes model with different emphasis, and construction algorithm is carrying out the same of reflection separation
When, different degrees of remains the unrecognized reflection in part, is lost information required for itself.Meanwhile such method is not
It is reflected suitable for the common double-deck ghost image.
Summary of the invention
The purpose of the present invention is to provide structure, relatively easy and effects that one kind can retain original image as far as possible
It is preferably based on the ghost image reflection minimizing technology of image gradient sparsity.
This ghost image based on image gradient sparsity provided by the invention reflects minimizing technology, includes the following steps:
S1. ghost image reflection model is constructed；
S2. the ghost image reflection model that step S1 is obtained is converted into ghost image reflection Variation Model；
S3. the ghost image reflection Variation Model that step S2 is obtained is solved；
S4. the model result obtained to step S3, is exported in a manner of picture, after obtaining final ghost image reflection removal
Image.
Ghost image reflection model described in step S1, specially using following formula as ghost image reflection model:
G=u+Hs
G is observed image and g ∈ R in formulaMN×1, u is the image and u ∈ R after ghost image reflection removalMN×1, s is reflected image
And s ∈ RMN×1, H is the double-deck reflection convolution matrix and H ∈ RMN×MN。
The ghost image reflection model that step S1 is obtained is converted into ghost image reflection Variation Model described in step S2, is specially adopted
Reflected image is separated with the method for minimizing total variation, to obtain ghost image reflection Variation Model.
The ghost image reflects Variation Model, specially reflects Variation Model as ghost image using following formula:
In formulaFor the minimum value for seeking u and s,For p norm, 0 < p < 1, RxFor the difference in image level direction
Operator matrix, RyFor the difference operator matrix of image vertical direction, u is the image and u ∈ R after ghost image reflection removalMN×1, s is
Reflected image and s ∈ RMN×1, u=g-Hs.
The ghost image reflection Variation Model that step S2 is obtained is solved described in step S3, is specially weighted using iteration
Least square method solves the ghost image reflection Variation Model that step S2 is obtained.
The ghost image reflection Variation Model obtained to step S2 solves, and is specially asked using following steps
Solution:
A. the u=g-Hs ghost image reflection Variation Model for bringing step S2 into is obtained:
In formulaFor the minimum value for seeking u and s,For p norm, 0 < p < 1, RxFor the difference in image level direction
Operator matrix, RyFor the difference operator matrix of image vertical direction, u is the image and u ∈ R after ghost image reflection removalMN×1, s is
Reflected image and s ∈ RMN×1, u=g-Hs；
B. the variable in step A variable replacement is carried out to obtain:
Z in formula1=Rx(g-Hs), Z2=Ry(g-Hs), Z3=RxS, Z4=Rys；
C. according to the definition of p norm, by the model conversion of step B are as follows:
W is diagonal matrix and its diagonal element in formulaε is positive parameter；
D. iteration weighted least-squares method is used, the model of step C is solved, obtains the optimality condition of model such as
Shown in following formula:
E. to above-mentioned Solving Linear, to obtain optimal solution s；
F. the image u after ghost image reflection removal is obtained using formula u=g-Hs.
This ghost image based on image gradient sparsity provided by the invention reflects minimizing technology, to anti-containing the double-deck ghost image
The translation distance and light attenuation coefficient penetrated between two layers of reflection of image are estimated, obtain reflection balancing matrix, then to anti-
It penetrates disjunctive model to obtain optimizing equation using image gradient sparsity, obtains transmission image finally by optimization equation is solved
With an approximate evaluation of reflected image；Therefore the method for the present invention can largely contract while removing image reflection
The short processing time of algorithm, retain the structure of image as far as possible, the detailed information such as texture are preferably comprehensive to obtain
Energy.
Detailed description of the invention
Fig. 1 is the method flow schematic diagram of the method for the present invention.
Fig. 2 is the contrast effect schematic diagram of the embodiment one of the method for the present invention.
Fig. 3 is the contrast effect schematic diagram of the embodiment two of the method for the present invention.
Specific embodiment
It is as shown in Figure 1 the method flow schematic diagram of the method for the present invention: provided by the invention this dilute based on image gradient
The ghost image for dredging property reflects minimizing technology, includes the following steps:
S1. ghost image reflection model is constructed；
In daily life, when across layer of transparent glass photographed, that there are glass is anti-for the picture that usually takes
It penetrates, and has second layer reflection and the problem of translation is with decaying occurs on the basis of first layer reflection；Assuming that translation distance with
And therefore attenuation by optimizing it is known that can obtain required removing reflected image；
According to the uniqueness that such is reflected, available following reflection model
G ∈ R in formulaM×N, u ∈ RM×N, s ∈ RM×N, the double-deck reflection convolution kernel of h expression.In actual process, above formula can
Be depicted in the form of matrix-vector come then the above reflection model can be converted into
G=u+Hs
G is observed image and g ∈ R in formulaMN×1, u is the image and u ∈ R after ghost image reflection removalMN×1, s is reflected image
And s ∈ RMN×1, H is the double-deck reflection convolution matrix and H ∈ RMN×MN；
S2. the ghost image reflection model that step S1 is obtained is converted into ghost image reflection Variation Model, specially using minimum
Total variation algorithm separates reflected image；
In the specific implementation, the reflection model as obtained in step S1 is ill, and the solution for meeting the equation has very
It is a variety of；But it can be found that having, the total variation of the image of reflection is bigger than the total variation of unreflected image, thus image gradient has
Sparsity；Therefore reflection can be efficiently separated using the method for minimizing total variation；Reflection model can be converted into step S1
Following Variation Model:
In formulaFor the minimum value for seeking u and s,For p norm, 0 < p < 1, RxFor the difference in image level direction
Operator matrix, RyFor the difference operator matrix of image vertical direction, u is the image and u ∈ R after ghost image reflection removalMN×1, s is
Reflected image and s ∈ RMN×1, u=g-Hs；
S3. the ghost image reflection Variation Model that step S2 is obtained is solved, specially uses iteration weighted least-squares
Method solves the ghost image reflection Variation Model that step S2 is obtained；
A. the u=g-Hs ghost image reflection Variation Model for bringing step S2 into is obtained:
In formulaFor the minimum value for seeking u and s,For p norm, 0 < p < 1, RxFor the difference in image level direction
Operator matrix, RyFor the difference operator matrix of image vertical direction, u is the image and u ∈ R after ghost image reflection removalMN×1, s is
Reflected image and s ∈ RMN×1, u=g-Hs；
B. the variable in step A variable replacement is carried out to obtain:
Z in formula1=Rx(g-Hs), Z2=Ry(g-Hs), Z3=RxS, Z4=Rys；
C. according to the definition of p norm, have for vector x:
Wherein W is diagonal matrix, and its diagonal element is
Therefore, by the model conversion of step B are as follows:
W is diagonal matrix and its diagonal element in formulaε is positive parameter, ε=1 herein
×10-3；
D. iteration weighted least-squares method is used, the model of step C is solved, obtains the optimality condition of model such as
Shown in following formula:
E. to above-mentioned Solving Linear, to obtain optimal solution s；
In the specific implementation, it needs to repeat step A~E more times, can just obtain ideal effect；
F. the image u after ghost image reflection removal is obtained using formula u=g-Hs；
S4. the model result obtained to step S3, is exported in a manner of picture, after obtaining final ghost image reflection removal
Image.
As shown in Figures 2 and 3, be the effect diagram of the method for the present invention: Fig. 2-1 is original image, and Fig. 2-2 is to use this
Literary method reflected image in p=0.2, Fig. 2-3 are the image that reflection is gone using this paper inventive method；Fig. 2-4 is Shih
Method output reflected image, Fig. 2-5 be Shih method export removes reflected image.Fig. 3-1 is original image, Fig. 3-2
To use context of methods reflected image in p=0.55, Fig. 3-3 is the image that reflection is gone using this paper inventive method.Figure
The reflected image that the method that 3-4 is Shih exports, what the method that Fig. 3-5 is Shih exported removes reflected image.In addition, side of the present invention
The method of method and Shih, in the original image in processing Fig. 2 and Fig. 3, the processing time comparison of the two is as shown in table 1 below:
Table 1 handles time comparison schematic table
| The time-consuming of this patent method | The time-consuming of the method for Shih | |
| Fig. 2 | 1784.47 | 13796.46 |
| Fig. 3 | 158.83 | 1132.34 |
By Fig. 2, Fig. 3 and table 1, it can be seen that, the method for the present invention is no matter on treatment effect, or handles in time-consuming,
Better than existing method.
Claims (6)
1. a kind of ghost image based on image gradient sparsity reflects minimizing technology, include the following steps:
S1. ghost image reflection model is constructed；
S2. the ghost image reflection model that step S1 is obtained is converted into ghost image reflection Variation Model；
S3. the ghost image reflection Variation Model that step S2 is obtained is solved；
S4. the model result obtained to step S3, is exported in a manner of picture, the figure after obtaining final ghost image reflection removal
Picture.
2. the ghost image according to claim 1 based on image gradient sparsity reflects minimizing technology, it is characterised in that step
Ghost image reflection model described in S1, specially using following formula as ghost image reflection model:
G=u+Hs
G is observed image and g ∈ R in formulaMN×1, u is the image and u ∈ R after ghost image reflection removalMN×1, s is reflected image and s
∈RMN×1, H is the double-deck reflection convolution matrix and H ∈ RMN×MN。
3. the ghost image according to claim 1 or 2 based on image gradient sparsity reflects minimizing technology, it is characterised in that step
The ghost image reflection model that step S1 is obtained is converted into ghost image reflection Variation Model described in rapid S2, it is specially total using minimizing
Variational algorithm separates reflected image, to obtain ghost image reflection Variation Model.
4. the ghost image according to claim 3 based on image gradient sparsity reflects minimizing technology, it is characterised in that described
Ghost image reflect Variation Model, specially using following formula as ghost image reflection Variation Model:
In formulaFor the minimum value for seeking u and s,For p norm, 0 < p < 1, RxFor the difference operator in image level direction
Matrix, RyFor the difference operator matrix of image vertical direction, u is the image and u ∈ R after ghost image reflection removalMN×1, s is reflection
Image and s ∈ RMN×1, u=g-Hs.
5. the ghost image according to claim 1 or 2 based on image gradient sparsity reflects minimizing technology, it is characterised in that step
The ghost image reflection Variation Model that step S2 is obtained is solved described in rapid S3, specially uses iteration weighted least-squares method
The ghost image reflection Variation Model that step S2 is obtained is solved.
6. the ghost image according to claim 5 based on image gradient sparsity reflects minimizing technology, it is characterised in that described
The ghost image reflection Variation Model that step S2 is obtained solve, specially solved using following steps:
A. the u=g-Hs ghost image reflection Variation Model for bringing step S2 into is obtained:
In formulaFor the minimum value for seeking u and s,For p norm, 0 < p < 1, RxFor the difference operator in image level direction
Matrix, RyFor the difference operator matrix of image vertical direction, u is the image and u ∈ R after ghost image reflection removalMN×1, s is reflection
Image and s ∈ RMN×1, u=g-Hs；
B. the variable in step A variable replacement is carried out to obtain:
Z in formula1=Rx(g-Hs), Z2=Ry(g-Hs), Z3=RxS, Z4=Rys；
C. according to the definition of p norm, by the model conversion of step B are as follows:
W is diagonal matrix and its diagonal element in formulaε is positive parameter；
D. iteration weighted least-squares method is used, the model of step C is solved, the optimality condition such as following formula of model is obtained
It is shown:
E. to above-mentioned Solving Linear, to obtain optimal solution s；
F. the image u after ghost image reflection removal is obtained using formula u=g-Hs.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910276482.6A CN109978798B (en) | 2019-04-08 | 2019-04-08 | Ghost reflection removing method based on image gradient sparsity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910276482.6A CN109978798B (en) | 2019-04-08 | 2019-04-08 | Ghost reflection removing method based on image gradient sparsity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109978798A true CN109978798A (en) | 2019-07-05 |
| CN109978798B CN109978798B (en) | 2022-11-29 |
Family
ID=67083323
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910276482.6A Expired - Fee Related CN109978798B (en) | 2019-04-08 | 2019-04-08 | Ghost reflection removing method based on image gradient sparsity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109978798B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112102182A (en) * | 2020-08-31 | 2020-12-18 | 华南理工大学 | Single image reflection removing method based on deep learning |
| CN113379608A (en) * | 2020-03-10 | 2021-09-10 | Tcl科技集团股份有限公司 | Image processing method, storage medium and terminal equipment |
| US11528435B2 (en) | 2020-12-25 | 2022-12-13 | Industrial Technology Research Institute | Image dehazing method and image dehazing apparatus using the same |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014072387A2 (en) * | 2012-11-09 | 2014-05-15 | Shell Internationale Research Maatschappij B.V. | Method for deghosting seismic data acquired by a marine seismic source and receiver assembly |
| CN106651798A (en) * | 2016-12-16 | 2017-05-10 | 四川长虹电器股份有限公司 | Effective image mirror reflection removal method |
| CN107016653A (en) * | 2017-03-29 | 2017-08-04 | 中国人民解放军信息工程大学 | Combine the CT image sparse angle reconstruction method and devices of total variation based on total curvature |
| WO2018099321A1 (en) * | 2016-11-30 | 2018-06-07 | 华南理工大学 | Generalized tree sparse-based weighted nuclear norm magnetic resonance imaging reconstruction method |
-
2019
- 2019-04-08 CN CN201910276482.6A patent/CN109978798B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014072387A2 (en) * | 2012-11-09 | 2014-05-15 | Shell Internationale Research Maatschappij B.V. | Method for deghosting seismic data acquired by a marine seismic source and receiver assembly |
| WO2018099321A1 (en) * | 2016-11-30 | 2018-06-07 | 华南理工大学 | Generalized tree sparse-based weighted nuclear norm magnetic resonance imaging reconstruction method |
| CN106651798A (en) * | 2016-12-16 | 2017-05-10 | 四川长虹电器股份有限公司 | Effective image mirror reflection removal method |
| CN107016653A (en) * | 2017-03-29 | 2017-08-04 | 中国人民解放军信息工程大学 | Combine the CT image sparse angle reconstruction method and devices of total variation based on total curvature |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113379608A (en) * | 2020-03-10 | 2021-09-10 | Tcl科技集团股份有限公司 | Image processing method, storage medium and terminal equipment |
| CN112102182A (en) * | 2020-08-31 | 2020-12-18 | 华南理工大学 | Single image reflection removing method based on deep learning |
| US11528435B2 (en) | 2020-12-25 | 2022-12-13 | Industrial Technology Research Institute | Image dehazing method and image dehazing apparatus using the same |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109978798B (en) | 2022-11-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110570371B (en) | Image defogging method based on multi-scale residual error learning | |
| CN109978798A (en) | Ghost image based on image gradient sparsity reflects minimizing technology | |
| CN106204449B (en) | A kind of single image super resolution ratio reconstruction method based on symmetrical depth network | |
| CN111062892B (en) | Single image rain removing method based on composite residual error network and deep supervision | |
| CN110570363A (en) | Image defogging method based on Cycle-GAN with pyramid pooling and multi-scale discriminator | |
| CN109087269A (en) | Low light image Enhancement Method and device | |
| CN106157268B (en) | One kind being based on the convex approximate degraded image restored method of L0 | |
| CN108564549A (en) | A kind of image defogging method based on multiple dimensioned dense connection network | |
| CN102243711A (en) | Neighbor embedding-based image super-resolution reconstruction method | |
| CN110648316B (en) | Steel coil end face edge detection method based on deep learning | |
| CN104504658A (en) | Single image defogging method and device on basis of BP (Back Propagation) neural network | |
| CN107194380A (en) | The depth convolutional network and learning method of a kind of complex scene human face identification | |
| CN111161360A (en) | Retinex theory-based image defogging method for end-to-end network | |
| CN113850324B (en) | Multispectral target detection method based on Yolov4 | |
| CN110634103A (en) | Image demosaicing method based on generation of countermeasure network | |
| CN107239781A (en) | A kind of super spectral reflectivity method for reconstructing based on RGB image | |
| CN103839236A (en) | Image white balance method based on sparse representation | |
| CN112669249A (en) | Infrared and visible light image fusion method combining improved NSCT (non-subsampled Contourlet transform) transformation and deep learning | |
| CN116343015A (en) | Medical food water content measurement system based on artificial intelligence | |
| CN115311186A (en) | Cross-scale attention confrontation fusion method for infrared and visible light images and terminal | |
| CN112164010A (en) | Multi-scale fusion convolution neural network image defogging method | |
| Hsu et al. | Object detection using structure-preserving wavelet pyramid reflection removal network | |
| Khan et al. | Deep image enhancement for ill light imaging | |
| CN115063833A (en) | Machine room personnel detection method based on image layered vision | |
| CN113205462B (en) | Photon reflectivity image denoising method based on neural network learning prior |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20221129 |