CN106373102B - Single image rain removing method based on gradient domain analysis - Google Patents
Single image rain removing method based on gradient domain analysis Download PDFInfo
- Publication number
- CN106373102B CN106373102B CN201610806663.1A CN201610806663A CN106373102B CN 106373102 B CN106373102 B CN 106373102B CN 201610806663 A CN201610806663 A CN 201610806663A CN 106373102 B CN106373102 B CN 106373102B
- Authority
- CN
- China
- Prior art keywords
- gradient
- rain
- texture
- high frequency
- domain analysis
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 20
- 239000011159 matrix material Substances 0.000 claims abstract description 5
- 238000003384 imaging method Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 2
- 238000004800 variational method Methods 0.000 claims description 2
- 230000002146 bilateral effect Effects 0.000 claims 1
- 230000015556 catabolic process Effects 0.000 abstract 1
- 238000012545 processing Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000035807 sensation 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention is the single image rain removing method based on gradient domain analysis, and rain usually shows nicking shape in the picture, thus is in different statistical properties to the gradients affect of X-direction and Y-direction.Based on this phenomenon, proposes and a kind of rain frame is gone based on gradient domain analysis.Rain problem is gone to be intended to restore the texture of image, theoretically texture is solved by the gradient of mutually perpendicular direction.If there is a direction, gradient interference is minimum, goes rain problem that can be simplified to solve the gradient of its vertical direction.The direction of most rain lines, it is exactly that gradient interferes the smallest direction, the present invention determines this direction by calculating the histograms of oriented gradients (HOG) of image different zones, and proposing a kind of picture breakdown frame based on the full variation of anisotropy and matrix rank minimization, the gradient being disturbed, which is resolved into, indicates texture and expression rain line two parts.Compared with traditional method based on study, what is newly proposed goes rain frame speed to improve 60 times, while preferable effect can be obtained.
Description
Technical field
Digital Image Processing and computer vision.The present invention is that a kind of single image based on gradient domain analysis goes to rain side
Method, it is intended to which how research eliminates the interference of rain line using image self information, to restore original texture.Different to video sequence
Column processing lacks the auxiliary of before and after frames, and make single image goes rain to become more challenging.
Background technique
Rain is boisterous one kind, in highlighted striated in imaging process, scene information is blocked, is produced new
Gradient information.Thus, visible sensation method of the rainy day based on characteristics of image such as identifies, detection etc. can usually fail.The research of early stage
It is concerned with how processing video, (Same Scene is not hidden in some frames by means of the relevance in video on the content frame of front and back
Gear), restore capped image texture.However, single image goes rain problem, lacks the auxiliary of context, have more
Challenge.
The rain removing method of single image can be divided into 3 kinds, based on the method for training set study, the method based on filtering, and base
In the method for low-rank, respectively there are advantage and disadvantage.Method based on study can efficiently separate open rain line, the disadvantage is that time-consuming more long, limit
Its application has been made, and when texure information is similar with rain line, has hardly resulted in effective dictionary in training process, leads to this
Class texture information is accidentally rejected.Method based on filtering and low-rank, time-consuming is few, the disadvantage is that robustness is not high.
Summary of the invention
The present invention is directed to design a kind of efficient and effective single image rain removing method.Rain usually shows perpendicular in the picture
Striated has directionality, thus is in different statistical properties to the gradients affect of X-direction and Y-direction.Based on this phenomenon, this hair
Bright propose a kind of removes rain frame based on gradient domain analysis.
Rain problem is gone to be intended to restore the texture of image, theoretically texture can be by Poisson's equation by orthogonal two sides
To gradient solve.Thus, if there is a direction, minimum is interfered in the gradient that this side up, can be ignored, remove rain
Problem can be simplified to solve the gradient information in another direction.Original image I is first divided into high-frequency I with two-sided filter by the present inventionHF
With low frequency ILFTwo parts make almost all of rain line information be retained in high frequency section, thus we only handle IHF.It is most
The direction of rain line is exactly that gradient interferes the smallest direction, the histograms of oriented gradients that the present invention passes through calculating image different zones
(HOG) this direction is determined, calculation formula is as follows:
It is region IiWeight, α (Ii) it is region IiThe direction estimated with HOG.Due to not
With the background color in region, the difference of illumination, rain is different in the annoyance level that different zones are presented, and just seeks one using weighting method
The best direction of a fitting.It is with image entropy eiIt seeks.
After obtaining the smallest direction β of gradient interference, we can seek the gradient in the direction β, and the ladder perpendicular to the direction β
Degree.The former goes rain problem how to be converted by the latter not by the interference of rainResolve into expression textureWith expression rain
LineTwo parts,
Matrix decomposition frame based on the full variation of anisotropy and matrix rank minimization, will be perpendicular to the gradient in the direction β
Being decomposed into indicates textureWith expression rain lineTwo parts, rain line has similar characteristic in an image scene, thus uses
Low-rank constraint, the anisotropic full variational methods of texture information obtain
Further consider imaging noise, can be converted into
(3) in formula first two be full variation materialization, Section 3 is imaging noise, and Section 4 is the specific of low-rank constraint
Change.μ and υ is scalar, and the value range of ρ is (0,1), with this energy minimization problem, how to be solvedWithIt givesJust
Beginningization can be calculated with following formula
It is knownIt can be calculated with following formula
It obtainsAnd the gradient in the direction βLater, interference-free high frequency section is solved with Poisson's equation
Wherein
Finally rain result is gone to be represented byThe present invention is compared with traditional method based on study, speed
Degree improves 60 times, while being able to maintain more image detail informations.
Claims (1)
1. a kind of single image rain removing method based on gradient domain analysis, which is characterized in that the described method includes:
Original image is resolved into high frequency section I with bilateral filteringHFWith low frequency part ILF, make the rainy texture information generated all protect
Stay in high frequency section；
A direction is found with histograms of oriented gradients (HOG), is disturbed the gradient of high frequency imaging part most in this direction
It is few, it can be ignored, wherein according to formulaDetermine that gradient interferes the smallest direction β, ω (Ii) be
Region IiWeight, α (Ii) it is region IiHOG estimation direction；
Matrix decomposition frame based on the full variation of anisotropy and matrix rank minimization, will be perpendicular to the gradient in the direction βIt decomposes
For at expression textureWith expression rain lineTwo parts constrain rain line with low-rankWith anisotropic full variational methods line
ReasonTexture is calculatedGradient interferes the gradient of the smallest direction βIt is solved without interruption by Poisson's equation
High frequency sectionI.e. according to Poisson's equationSolve interference-free high frequency sectionWherein
According to formulaRain result is gone in determination.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610806663.1A CN106373102B (en) | 2016-09-07 | 2016-09-07 | Single image rain removing method based on gradient domain analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610806663.1A CN106373102B (en) | 2016-09-07 | 2016-09-07 | Single image rain removing method based on gradient domain analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106373102A CN106373102A (en) | 2017-02-01 |
| CN106373102B true CN106373102B (en) | 2019-09-10 |
Family
ID=57898900
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610806663.1A Active CN106373102B (en) | 2016-09-07 | 2016-09-07 | Single image rain removing method based on gradient domain analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106373102B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103310428A (en) * | 2012-03-08 | 2013-09-18 | 财团法人工业技术研究院 | Method and device for removing rainprint in image based on single image |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8879855B2 (en) * | 2012-08-17 | 2014-11-04 | Nec Laboratories America, Inc. | Image segmentation for large-scale fine-grained recognition |
-
2016
- 2016-09-07 CN CN201610806663.1A patent/CN106373102B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103310428A (en) * | 2012-03-08 | 2013-09-18 | 财团法人工业技术研究院 | Method and device for removing rainprint in image based on single image |
Non-Patent Citations (3)
| Title |
|---|
| Fast Algorithms for Recovering a Corrupted Low-Rank Matrix;Arvind Ganesh等;《2009 3rd IEEE International Workshop on Computational Advances in Multi-Sensor Adaptive Processing》;20091216;第213-216页 |
| 卡通纹理分解和全变分梯度算法实现图像恢复;蒋正金等;《计算机工程与应用》;20140115;第50卷(第2期);第162-169页 |
| 基于求解泊松方程和梯度的图像修复的研究;杨勇等;《计算机技术与发展》;20080210;第18卷(第2期);正文第98-100页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106373102A (en) | 2017-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105654436B (en) | A kind of backlight image enhancing denoising method based on prospect background separation | |
| Zheng et al. | Single-image-based rain and snow removal using multi-guided filter | |
| Huang et al. | An efficient visibility enhancement algorithm for road scenes captured by intelligent transportation systems | |
| Guo et al. | LIME: Low-light image enhancement via illumination map estimation | |
| Guo | LIME: A method for low-light image enhancement | |
| Chen et al. | Visual depth guided color image rain streaks removal using sparse coding | |
| Ju et al. | Single image dehazing via an improved atmospheric scattering model | |
| Li et al. | SURE-based optimization for adaptive sampling and reconstruction | |
| CN105957030B (en) | One kind being applied to the enhancing of thermal infrared imager image detail and noise suppressing method | |
| Wang et al. | A fast single-image dehazing method based on a physical model and gray projection | |
| CN106651871B (en) | Automatic filling method for depth image hole | |
| CN103686194A (en) | Video denoising method and device based on non-local mean value | |
| US20150278605A1 (en) | Apparatus and method for managing representative video images | |
| Fang et al. | Single image dehazing and denoising with variational method | |
| CN105809673B (en) | Video foreground dividing method based on SURF algorithm and the maximum similar area of merging | |
| CN103886553A (en) | Method and system for non-local average value denoising of image | |
| CN102800055B (en) | Low-order decomposition method for blind deblurring of images | |
| CN102800054A (en) | Image blind deblurring method based on sparsity metric | |
| Kaur et al. | Image enhancement techniques: A selected review | |
| Zhang et al. | An improved watershed algorithm for color image segmentation | |
| CN106373102B (en) | Single image rain removing method based on gradient domain analysis | |
| CN112750089A (en) | Optical remote sensing image defogging method based on local block maximum and minimum pixel prior | |
| Wu et al. | Non-uniform low-light image enhancement via non-local similarity decomposition model | |
| CN117635471A (en) | Low-illumination image enhancement method for wireless capsule endoscope | |
| Li et al. | Image de-noising of locust sections based on adaptive wavelet and partial differential equation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |