CN103514582A - Visual saliency-based image deblurring method - Google Patents
Visual saliency-based image deblurring method Download PDFInfo
- Publication number
- CN103514582A CN103514582A CN201210214195.0A CN201210214195A CN103514582A CN 103514582 A CN103514582 A CN 103514582A CN 201210214195 A CN201210214195 A CN 201210214195A CN 103514582 A CN103514582 A CN 103514582A
- Authority
- CN
- China
- Prior art keywords
- image
- saliency
- conspicuousness
- fuzzy core
- picture
- 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
Images
Abstract
The invention relates to a visual saliency-based image deblurring method. The method is characterized in that according to a visual attention mechanism, saliency dividing is carried out on a to-be-processed image by utilizing image sharpness as a saliency condition, wherein the saliency dividing includes division of a saliency portion and a non-saliency portion; blurring kernel estimation is carried out on the non-saliency portion so as to obtain a blurring kernel K of the image; and a compensation mechanism is introduced and thus the obtained blurring kernel is applied to the whole to-be-processed image, and a ringing-free effect occurs at the edge of the saliency portion so as to obtain an image processed by deblurring. According to the method provided by the invention, as the visual attention-based saliency study, the blurred image is divided into two portions including the saliency portion and the non-saliency portion based on a saliency map; only the blurring kernel of the non-saliency portion is estimated and that is, the blurred portion of the blurred image is processed so as to maintain the clear portion. Therefore, the influence on blurred images with different blurring degrees of different portions and different blurring directions due to the same blurring kernel estimation processing can be eliminated.
Description
Technical field
The present invention relates to the deblurring method in a kind of image restoration field, specifically relate to a kind of single image that is applied in, and the image deblurring method of the conspicuousness research based on vision noticing mechanism, can efficiently realize rapidly the deblurring of blurred picture.
Background technology
In single image deblurring technology, blurred picture has comprised fuzzy caused image degradation part incessantly, also comprises the image degradation part that noise causes.Therefore, first removing noise is exactly a large difficult point, because the detailed information of image is covered by noise, rashly removing noise process will inevitably damage the detailed information of image, causes the serious degradation of final solving result; Meanwhile, the deblurring based on single image also will face two problems---estimation and a high-quality image deconvolution process of point spread function (fuzzy core).According to experiment, show, very perfect even if point spread function (fuzzy core) has been assessed, if be directly used for deconvoluting, also can cause very serious ringing; When image blurring trajectorie is everywhere different, also can cause deblurring poor effect simultaneously.Therefore how realizing quickly and efficiently the deblurring of image, have good deblurring effect simultaneously, is a problem that is worth research.
Single image deblurring algorithm is a kind of problem of pathosis.Because our known information is not sufficient to make us to obtain accurate result.For realizing quickly and efficiently the deblurring of image, had a lot of methods for motion blur to some restrictive hypothesis, for example suppose that motion blur is parameterized linear movement, in technology, there is now following several method:
1) suppose that motion blur is isotropic;
2) utilize the variational Bayesian method of natural image statistics;
3) fuzzy core that the fuzzy message of utilizing alpha matting to obtain has removed to estimate each pixel;
4) the alpha matte that utilizes transparency (transparency) to change carrys out ambiguous estimation core.
5) utilize a series of optimisation technique to accomplish good deblurring algorithm.
Although these above-mentioned methods have also realized the deblurring of image, guaranteeing, in universality and high efficiency situation, also to exist suitable defect.
Summary of the invention
The object of this invention is to provide a kind of in the situation that not affecting image deblurring quality, effectively reduce working time and have universality based on the significant image deblurring method of vision.
For achieving the above object, the present invention is by the following technical solutions:
The significant image deblurring method of vision, according to vision noticing mechanism, first utilizes whether image is clear carries out conspicuousness division for conspicuousness condition to pending image, i.e. conspicuousness part and non-conspicuousness part; Then, non-conspicuousness is partly carried out to the fuzzy core that fuzzy core estimates to obtain image
k; Finally, introduce compensation mechanism, the fuzzy core obtaining is applied in the pending image of view picture, make the edge of conspicuousness part without ringing effect simultaneously, and then obtain image after deblurring.
Non-conspicuousness is partly carried out to the fuzzy core that fuzzy core estimates to obtain image
kcomprise the following steps:
1. prediction: first calculate true picture
ledge
xwith
ydirection gradient mapping
p x , P y , to predict the true picture after carrying out smooth region inhibition noise
lprominent edge; Wherein, the true picture of primary iteration
lby given blurred picture
bobtain the true picture except primary iteration
l, be the true picture obtaining after the step of deconvoluting in a front iteration
l;
2. fuzzy core
kestimate: the gradient mapping after utilization prediction
p x , P y and given blurred picture
bgradient shine upon ambiguous estimation core
k;
3. deconvolute: by the fuzzy core of gained
kwith given blurred picture
bobtain true picture
l.
Described prediction steps comprises two-sided filter to be processed, impulse filter process and gradient magnitude thresholding processing procedure; First at true picture
lupper utilization two-sided filter is to reduce noise and little details; Then use impulse wave filter to true picture
lcarry out sharpened edge recovery.
Obtaining the most coarse fuzzy core
kstage, to given blurred picture
bcarry out down-sampled to start prediction steps; And obtaining meticulousr fuzzy core
kstage, the true picture that coarse stage is obtained
lby bilinear interpolation mode, carry out rising sampling, to carry out the iteration in meticulousr stage.
Adopt the present invention of technique scheme, conspicuousness research based on vision attention, by its remarkable figure, blurred picture is divided into two parts, conspicuousness part and non-conspicuousness part, only estimate the fuzzy core of non-conspicuousness part, only the fuzzy part of blurred picture is processed, kept its part comparatively clearly.So just can avoid blurred picture because of the fog-level of different piece and the difference of blur direction, and with same ambiguous estimation core, process the impact bringing.In addition, in the non-conspicuousness part of image fuzzy core estimating step, introduce forecasting mechanism and blurred picture is carried out to two-sided filter, impulse wave filter and gradient magnitude thresholding process, the sharp keen property at Recovery image edge and suppress noise at smooth region, reduces ringing as far as possible.In addition, when estimating for fuzzy core, by utilizing image gradient rather than pixel value to obtain the effect of acceleration, experiment shows, this algorithm can improve a lot shortening on the processing time.And introducing compensation mechanism, make original blurred picture be converted into space invariance motion blur by the constant motion blur of non-space, the fuzzy core of prediction can be used in entire image, make the edge of conspicuousness part without ringing effect simultaneously, and then can carry out deblurring operation to entire image.
Accompanying drawing explanation
Fig. 1 is the image deblurring algorithm flow chart based on conspicuousness of the present invention.
Fig. 2 is quick deblurring algorithm block diagram in the present invention.
Fig. 3 is original image in experimental result of the present invention.
Fig. 4 significantly schemes in experimental result of the present invention.
Fig. 5 is the overall fuzzy graph obtaining by compensation mechanism in experimental result of the present invention.
Fig. 6 is the result figure after deblurring in experimental result of the present invention.
Embodiment
The present invention is according to vision noticing mechanism, first utilizes whether image is clear carries out conspicuousness division for conspicuousness condition to pending image, obtains conspicuousness part and non-conspicuousness part.In the present embodiment, adopt the conspicuousness extraction algorithm based on frequency field, obtain required Saliency maps.Why the present invention proposes based on the significant image deblurring method of vision, because the different piecemeals in image have difference on fog-level or the difference in blur direction, even if a block during next is fuzzy by difference has estimated a fuzzy core, utilize this fuzzy core to carry out deconvolution algorithm to entire image and also only can make the effect of not fuzzy part become poorer.
After conspicuousness is divided, can obtain removing the non-conspicuousness parts of images of conspicuousness part, then for this part, just can use quick deblurring algorithm to carry out fuzzy core estimation, and use when introducing compensation mechanism for conspicuousness part.Due to non-conspicuousness part fuzzy be space invariance and rectangular image be relatively good processing, therefore can manage herein the rectangle connected domain of this part maximum, and because this part fuzzy form is single, so the fuzzy core that can be estimated by maximum rectangle connected domain is completely used as the fuzzy core of whole non-conspicuousness part.
Carry out fuzzy core and estimate it is for meticulous fuzzy core progressively
k, following three steps of this algorithm iteration: prediction, fuzzy core estimate and deconvolute, as shown in Figure 2.This algorithm is a kind of blind deconvolution algorithm based on single image estimating motion fuzzy core and true picture, and has introduced forecasting mechanism and utilized image gradient rather than its pixel value.
In prediction steps, first calculate true picture
ledge
xwith
ydirection gradient mapping
p x , P y , to predict the true picture after carrying out smooth region inhibition noise
lprominent edge.Wherein, the true picture of primary iteration
lby given blurred picture
bobtain the true picture except primary iteration
l, be the true picture obtaining after the step of deconvoluting in a front iteration
l.In fuzzy core estimating step, the gradient mapping after utilization prediction
p x , P y and given blurred picture
bgradient shine upon ambiguous estimation core
k.In the step of deconvoluting, by the fuzzy core obtaining
kwith given blurred picture
bobtain true picture
l.And the true picture obtaining here
lto in the prediction steps of next iteration, apply to again.In order to make better effects and if ambiguous estimation core more efficiently
kand true picture
l, introduce the mechanism of a kind of " by coarse to meticulous ".When the most coarse stage, to given blurred picture
bcarry out down-sampled to start prediction steps.And obtain in coarse stage
l, by bilinear interpolation mode, carry out rising sampling, to carry out the iteration in meticulousr stage.
Prediction steps is comprising two-sided filter to be processed, impulse filter process and gradient magnitude thresholding processing procedure.First exist
lupper utilization two-sided filter is to reduce noise and little details.Then use impulse wave filter pair
lcarry out sharpened edge recovery.And obtain after shock filter
l 'comprising strengthen Liao edge and increasing noise, by calculating and thresholding
l 'gradient mapping
? x l ', y l ', and the mapping of gradient after blocking
p x , P y provided the end product of prediction steps.Introduce forecasting mechanism as far as possible Recovery image edge sharp keen property and at smooth region, suppress noise, reduce ringing.
When estimating for fuzzy core, by utilizing image gradient rather than pixel value to obtain the effect of acceleration, experiment shows, utilizes image gradient to shorten and processes required time.
Then introduce compensation mechanism, make original blurred picture be converted into space invariance motion blur by the constant motion blur of non-space, the fuzzy core of prediction can be used in entire image, make the edge of conspicuousness part without ringing effect simultaneously, and then can carry out deblurring operation to entire image.As shown in Figure 1, the fuzzy core of estimating according to non-conspicuousness part (background) is fuzzy by conspicuousness part (prospect), now be combined into a view picture blurred picture with non-conspicuousness part (background), and then by the fuzzy core of estimation, carry out de-convolution operation and obtain the image after deblurring.
Claims (4)
1. based on the significant image deblurring method of vision, it is characterized in that: according to vision noticing mechanism, first utilize whether image is clear carries out conspicuousness division for conspicuousness condition to pending image, i.e. conspicuousness part and non-conspicuousness part; Then, non-conspicuousness is partly carried out to the fuzzy core that fuzzy core estimates to obtain image
k; Finally, introduce compensation mechanism, the fuzzy core obtaining is applied in the pending image of view picture, make the edge of conspicuousness part without ringing effect simultaneously, and then obtain image after deblurring.
2. according to claim 1 based on the significant image deblurring method of vision, it is characterized in that: non-conspicuousness is partly carried out to the fuzzy core that fuzzy core estimates to obtain image
kcomprise the following steps:
1. prediction: first calculate true picture
ledge
xwith
ydirection gradient mapping
p x , P y , to predict the true picture after carrying out smooth region inhibition noise
lprominent edge; Wherein, the true picture of primary iteration
lby given blurred picture
bobtain the true picture except primary iteration
l, be the true picture obtaining after the step of deconvoluting in a front iteration
l;
2. fuzzy core
kestimate: the gradient mapping after utilization prediction
p x , P y and given blurred picture
bgradient shine upon ambiguous estimation core
k;
3. deconvolute: by the fuzzy core of gained
kwith given blurred picture
bobtain true picture
l.
3. according to claim 2 based on the significant image deblurring method of vision, it is characterized in that: described prediction steps comprises two-sided filter to be processed, impulse filter process and gradient magnitude thresholding processing procedure; First at true picture
lupper utilization two-sided filter is to reduce noise and little details; Then use impulse wave filter to true picture
lcarry out sharpened edge recovery.
4. according to claim 2 based on the significant image deblurring method of vision, it is characterized in that: obtaining the most coarse fuzzy core
kstage, to given blurred picture
bcarry out down-sampled to start prediction steps; And obtaining meticulousr fuzzy core
kstage, the true picture that coarse stage is obtained
lby bilinear interpolation mode, carry out rising sampling, to carry out the iteration in meticulousr stage.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210214195.0A CN103514582A (en) | 2012-06-27 | 2012-06-27 | Visual saliency-based image deblurring method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210214195.0A CN103514582A (en) | 2012-06-27 | 2012-06-27 | Visual saliency-based image deblurring method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103514582A true CN103514582A (en) | 2014-01-15 |
Family
ID=49897262
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210214195.0A Pending CN103514582A (en) | 2012-06-27 | 2012-06-27 | Visual saliency-based image deblurring method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103514582A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103985095A (en) * | 2014-05-20 | 2014-08-13 | 北京理工大学 | Image smoothing method based on significant structure metric model |
CN105574823A (en) * | 2015-12-14 | 2016-05-11 | 北京大学深圳研究生院 | Deblurring method and device for out-of-focus blurred image |
CN105574866A (en) * | 2015-12-15 | 2016-05-11 | 努比亚技术有限公司 | Image processing method and apparatus |
CN106097256A (en) * | 2016-05-31 | 2016-11-09 | 南京邮电大学 | A kind of video image fuzziness detection method based on Image Blind deblurring |
WO2017100971A1 (en) * | 2015-12-14 | 2017-06-22 | 北京大学深圳研究生院 | Deblurring method and device for out-of-focus blurred image |
CN107784631A (en) * | 2016-08-24 | 2018-03-09 | 中安消物联传感(深圳)有限公司 | Image deblurring method and device |
CN107870715A (en) * | 2016-09-27 | 2018-04-03 | 三星电子株式会社 | For providing the method and apparatus and recording medium of augmented reality image |
CN108510453A (en) * | 2018-03-07 | 2018-09-07 | 西安工程大学 | The intelligent traffic monitoring image deblurring method of view-based access control model attention mechanism |
CN109029270A (en) * | 2018-08-17 | 2018-12-18 | 浙江大丰实业股份有限公司 | Stage table top displacement monitoring platform |
CN110033415A (en) * | 2019-03-20 | 2019-07-19 | 东南大学 | A kind of image deblurring method based on Retinex algorithm |
CN110503611A (en) * | 2019-07-23 | 2019-11-26 | 华为技术有限公司 | The method and apparatus of image procossing |
CN110782401A (en) * | 2019-09-23 | 2020-02-11 | 浙江大华技术股份有限公司 | Image deblurring method and related device |
CN112215773A (en) * | 2020-10-12 | 2021-01-12 | 新疆大学 | Local motion deblurring method and device based on visual saliency and storage medium |
CN113313655A (en) * | 2021-06-28 | 2021-08-27 | 合肥工业大学 | Blind image deblurring method based on saliency mapping and gradient cepstrum technology |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101406041A (en) * | 2006-05-08 | 2009-04-08 | 三菱电机株式会社 | Method for reducing blur in an image of a scene and method for deblurring an image of a scene |
US20110229044A1 (en) * | 2010-03-16 | 2011-09-22 | Novatek Microelectronics Corp. | Hierarchical motion deblurring method for single image |
-
2012
- 2012-06-27 CN CN201210214195.0A patent/CN103514582A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101406041A (en) * | 2006-05-08 | 2009-04-08 | 三菱电机株式会社 | Method for reducing blur in an image of a scene and method for deblurring an image of a scene |
US20110229044A1 (en) * | 2010-03-16 | 2011-09-22 | Novatek Microelectronics Corp. | Hierarchical motion deblurring method for single image |
Non-Patent Citations (2)
Title |
---|
BING ZHOU 等: "Image Deblurring Based On Visual Saliency", 《 2012 INTERNATIONAL CONFERENCE ON SYSTEMS AND INFORMATICS (ICSAI)》, 20 May 2012 (2012-05-20), pages 1919 - 1922, XP032192901, DOI: doi:10.1109/ICSAI.2012.6223423 * |
杨勇杰: "运动模糊图像恢复方法", 《万方学位论文数据库》, 31 May 2010 (2010-05-31), pages 19 - 25 * |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103985095B (en) * | 2014-05-20 | 2016-08-24 | 北京理工大学 | Image smoothing method based on notable structure measurement model |
CN103985095A (en) * | 2014-05-20 | 2014-08-13 | 北京理工大学 | Image smoothing method based on significant structure metric model |
CN105574823A (en) * | 2015-12-14 | 2016-05-11 | 北京大学深圳研究生院 | Deblurring method and device for out-of-focus blurred image |
WO2017100971A1 (en) * | 2015-12-14 | 2017-06-22 | 北京大学深圳研究生院 | Deblurring method and device for out-of-focus blurred image |
CN105574823B (en) * | 2015-12-14 | 2018-10-09 | 北京大学深圳研究生院 | A kind of deblurring method and device of blurred picture out of focus |
CN105574866A (en) * | 2015-12-15 | 2016-05-11 | 努比亚技术有限公司 | Image processing method and apparatus |
CN106097256B (en) * | 2016-05-31 | 2019-01-01 | 南京邮电大学 | A kind of video image fuzziness detection method based on Image Blind deblurring |
CN106097256A (en) * | 2016-05-31 | 2016-11-09 | 南京邮电大学 | A kind of video image fuzziness detection method based on Image Blind deblurring |
CN107784631A (en) * | 2016-08-24 | 2018-03-09 | 中安消物联传感(深圳)有限公司 | Image deblurring method and device |
CN107784631B (en) * | 2016-08-24 | 2020-05-05 | 深圳市豪恩安全科技有限公司 | Image deblurring method and device |
CN107870715B (en) * | 2016-09-27 | 2021-05-18 | 三星电子株式会社 | Method and apparatus for providing augmented reality image and recording medium |
CN107870715A (en) * | 2016-09-27 | 2018-04-03 | 三星电子株式会社 | For providing the method and apparatus and recording medium of augmented reality image |
CN108510453A (en) * | 2018-03-07 | 2018-09-07 | 西安工程大学 | The intelligent traffic monitoring image deblurring method of view-based access control model attention mechanism |
CN108510453B (en) * | 2018-03-07 | 2021-06-29 | 西安工程大学 | Intelligent traffic monitoring image deblurring method based on visual attention mechanism |
CN109029270B (en) * | 2018-08-17 | 2020-11-27 | 浙江大丰实业股份有限公司 | Stage table displacement monitoring platform |
CN109029270A (en) * | 2018-08-17 | 2018-12-18 | 浙江大丰实业股份有限公司 | Stage table top displacement monitoring platform |
CN110033415A (en) * | 2019-03-20 | 2019-07-19 | 东南大学 | A kind of image deblurring method based on Retinex algorithm |
CN110503611A (en) * | 2019-07-23 | 2019-11-26 | 华为技术有限公司 | The method and apparatus of image procossing |
CN110782401B (en) * | 2019-09-23 | 2022-09-06 | 浙江大华技术股份有限公司 | Image deblurring method and related device |
CN110782401A (en) * | 2019-09-23 | 2020-02-11 | 浙江大华技术股份有限公司 | Image deblurring method and related device |
CN112215773A (en) * | 2020-10-12 | 2021-01-12 | 新疆大学 | Local motion deblurring method and device based on visual saliency and storage medium |
CN112215773B (en) * | 2020-10-12 | 2023-02-17 | 新疆大学 | Local motion deblurring method and device based on visual saliency and storage medium |
CN113313655A (en) * | 2021-06-28 | 2021-08-27 | 合肥工业大学 | Blind image deblurring method based on saliency mapping and gradient cepstrum technology |
CN113313655B (en) * | 2021-06-28 | 2022-09-23 | 合肥工业大学 | Blind image deblurring method based on saliency mapping and gradient cepstrum technology |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103514582A (en) | Visual saliency-based image deblurring method | |
JP5342068B2 (en) | Multiple frame approach and image upscaling system | |
CN107749987B (en) | Digital video image stabilization method based on block motion estimation | |
EP2164040B1 (en) | System and method for high quality image and video upscaling | |
Zheng et al. | Wavelet based nonlocal-means super-resolution for video sequences | |
CN105447828B (en) | The one-view image deblurring method of one-dimensional deconvolution is carried out along motion blur path | |
CN103310430A (en) | Method and apparatus for deblurring non-uniform motion blur | |
Vijayanagar et al. | Refinement of depth maps generated by low-cost depth sensors | |
CN108765288B (en) | POCS image super-resolution reconstruction method based on edge preservation | |
CN100525452C (en) | Frequency domain fast sub picture element global motion estimating method for image stability | |
CN110930324A (en) | Fuzzy star map restoration method | |
CN102750679A (en) | Blind deblurring method for image quality evaluation | |
CN102663675B (en) | Method and apparatus for scaling image processed by character superimposition | |
WO2015085637A1 (en) | Method for supplementarily drawing content at station logo region in video | |
CN111626943B (en) | Total variation image denoising method based on first-order forward and backward algorithm | |
CN110852947B (en) | Infrared image super-resolution method based on edge sharpening | |
CN105279742A (en) | Quick image denoising method on the basis of partition noise energy estimation | |
CN105825480A (en) | Image deblurring method based on sapphire surface defect automatic detection | |
Zhu et al. | Dast-net: Depth-aware spatio-temporal network for video deblurring | |
CN112069870A (en) | Image processing method and device suitable for vehicle identification | |
He et al. | Joint motion deblurring and superresolution from single blurry image | |
Xu et al. | Interlaced scan CCD image motion deblur for space-variant motion blurs | |
Zhang | Image motion deblurring based on salient structure selection and L0-2 norm kernel estimation | |
KR20130038495A (en) | Fast method for matching stereo images according to operation skip | |
Chen et al. | Video super-resolution using low rank matrix completion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20140115 |
|
RJ01 | Rejection of invention patent application after publication |