CN106530247B - A kind of multi-scale image restorative procedure based on structural information - Google Patents
A kind of multi-scale image restorative procedure based on structural information Download PDFInfo
- Publication number
- CN106530247B CN106530247B CN201610963234.5A CN201610963234A CN106530247B CN 106530247 B CN106530247 B CN 106530247B CN 201610963234 A CN201610963234 A CN 201610963234A CN 106530247 B CN106530247 B CN 106530247B
- Authority
- CN
- China
- Prior art keywords
- image
- scale
- kth
- layer
- reparation
- 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 description 15
- 230000008439 repair process Effects 0.000 claims abstract description 37
- 230000000694 effects Effects 0.000 claims abstract description 30
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 29
- 230000006870 function Effects 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims abstract description 13
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 9
- 238000012163 sequencing technique Methods 0.000 claims abstract description 5
- 238000005259 measurement Methods 0.000 claims abstract description 4
- 238000012913 prioritisation Methods 0.000 claims abstract description 4
- 238000001914 filtration Methods 0.000 claims description 35
- 238000005070 sampling Methods 0.000 claims description 7
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 238000009792 diffusion process Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000002146 bilateral effect Effects 0.000 description 2
- 238000012886 linear function Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect 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
Abstract
The present invention discloses a kind of multi-scale image reparation algorithm based on structural information, comprising: carries out multiple dimensioned decomposition to image I to be repaired, obtains a series of different scale tomographic images;Since out to out tomographic image, using the prioritization functions calculation method based on structural information, current kth scale tomographic image I is calculatedkThe priority of damaged area, sequencing according to priority, blocks and optimal matching blocks are found using the similarity measurement criterion that corresponding matching criterior is constituted, carry out the image repair of current kth scale layer, it obtains corresponding repairing effect figure and it is carried out to up-sample to rebuild to obtain the benchmark figure of -1 scale layer reparation of kth, the image for repairing -1 scale layer of kth obtains the preliminary reparation result under this scale;Step is tentatively repaired to result to compare with benchmark figure, obtains new -1 scale layer breakage image of kth;Obtain the final repairing effect figure of -1 scale layer of kth.Reparation sequence of the present invention is more accurate, and the picture structure texture after reparation is more naturally reasonable.
Description
Technical field
The invention belongs to computer pictures to restore field, be related to a kind of multi-scale image reparation calculation based on structural information
Method can be used for restoring rotten deterioration film, scratch elimination etc..
Background technique
Digital Image Inpainting is the process that information filling is carried out to loss of learning region in image to be repaired, reparation person
It needs to realize the recovery of breakage image using most appropriate method, while making observer that can not perceive what image was repaired
Trace guarantees that image reaches optimal artistic effect.
Classics are the reparation algorithms based on sample block that Criminisi et al. is proposed the most in digital picture reparation algorithm.
It determines the sequencing of reparation using priority size.Wherein priority is the product by confidence level item and data item.By
Cause to repair inaccuracy as the increase for repairing the number of iterations declines rapidly in confidence level, and its data item only considers isophote
Direction, structure description is excessively single, and it is inaccurate only to also result in result in a scale layer reparation.Domestic and foreign scholars propose thus
New priority calculates function and multiple dimensioned reparation scheduling algorithm is improved.
Summary of the invention
Deficiency existing for the image repair algorithm based on sample block for single scale, the present invention provide a kind of based on knot
The multi-scale image of structure information repairs algorithm, and the invention firstly uses weight navigational figure filtering WGIF methods to break to be repaired
Damage image carries out multiple dimensioned decomposition, obtains a series of scale layer;Then image repair is carried out since scale maximum layer,
During the image repair of each scale, the curvature of image and structural information are described into two characteristic informations and are introduced into priority
It calculates, the sequence for repairing filling is more reasonable.Technical scheme is as follows:
A kind of multi-scale image restorative procedure based on structural information, including the following steps:
1) multiple dimensioned decomposition is carried out to image I to be repaired, i.e., since the 0th layer of smallest dimension layer, using based on weight
The algorithm of guiding filtering is filtered to each scale tomographic image and down-sampling, is obtained using the filter window of M × N size a series of
Different scale tomographic image Ik;Wherein, (i, j) is pixel
Coordinate, m=[M/2], n=[N/2] are respectively represented no more than M/2, the maximum integer of N/2, WWGIF(u, v) is picture in filter window
The corresponding WGIF filtering weighting value of vegetarian refreshments (u, v);
2) since out to out layer, that is, top layer images, the prioritization functions calculation method P (p) based on structural information is utilized
=C (p) [wD·D(p)+wK·|K(p)|+wΓ·ΓI(p)] current kth scale tomographic image I, is calculatedkDamaged area it is preferential
Power;Wherein C (p) is the confidence level of point p, and D (p) is the data item of point p, and K (p) is curvature, ΓI(p) for structural information describe because
Son,
wD、wK、wΓFor weight, and meet:
3) sequencing according to priority finds Optimum Matching using the similarity measurement criterion that corresponding matching criterior is constituted
Block carries out the image repair of current kth scale layer, obtains corresponding repairing effect figure Iipt_k;
4) current kth scale layer image repair result up-sampling reconstruction is carried out to obtainAs -1 scale of kth
Layer repair benchmark figure, and according to step 2), 3) repair -1 scale layer of kth image Ik-1It obtains first under this scale
Step repairs result Iinitial_ipt_k-1;
5) the preliminary reparation result I that will be generated in step 4)initial_ipt_k-1With benchmark figureIt compares, obtains
To new -1 scale layer breakage image I of kthnew_k-1;
6) according to step 2) and 3) to -1 scale layer breakage image I of kth newly obtainednew_k-1Image repair is carried out, is obtained
The final repairing effect figure I of -1 scale layer of kthipt_k-1;
7) it repeats step 4), 5) and 6) successively traverses each scale layer, until completing the reparation of smallest dimension layer, then repair
Terminate.
Compared with the prior art, advantages of the present invention has following two points:
(1) present invention carries out multi-resolution decomposition to image by introducing weight guiding filtering, makes the edge and details of image
Information is protected well, facilitates the further reparation of image.
(2) present invention proposes a kind of priority calculation formula based on structural information, by curvature and has guarantor's local edge
Data item of the function as priority, keep the reparation sequence of image more reasonable, repairing effect is more natural.
It repairs filtering in algorithm in short, the present invention preferably resolves traditional multiscale transform information is caused largely to be lost and structure
Unreasonable problem is repaired, for repairing there is the image aspect of labyrinth to obtain preferable visual effect, reparation sequence is more
To be accurate, while having the characteristics that stronger robustness to various natural textures, has a wide range of applications.
Detailed description of the invention
A kind of multi-scale image based on structural information Fig. 1 of the invention repairs algorithm flow chart.
Fig. 2~4 are the repairing effect comparison that the present invention repairs algorithm to different images with traditional images, in which:
Figure (a) red area represents damaged area;
Scheme the repairing effect that (b) is classics Criminisi algorithm;
Scheming (c) is Entropy algorithm repairing effect;
Scheming (d) is the method for the present invention repairing effect
Specific embodiment
A kind of multi-scale image based on structural information of the present invention repairs algorithm, mainly consists of two parts: weight guidance
Filter the reparation of multi-resolution decomposition image, each scale layer.Concrete principle is as follows:
1: weight guiding filtering (Weighted Guided Image Filtering is abbreviated as WGIF) is a kind of guarantor edge
Image filter arithmetic, the performance for protecting edge are better than traditional bilateral filtering and guiding filtering.
The model of weight guiding filtering is derived from traditional guiding filtering model, in guiding filtering definition, uses local line
Property model, the certain point on such model hypothesis function and the point in its field are linear, this class model is highly suitable for
In the expression of non-analytic function.
According to navigational figure Filtering Model, it can be assumed that the filtering output q of this kind of two-dimensional function of image and input guide
Image I meets certain linear relationship in the two-dimentional window of restriction, specific as follows:
Wherein, (i, j) is that pixel is put centered on (x, y), any pixel in window N (x, y).To formula (1-1)
Gradient is sought simultaneously in both sides, available:
By formula (1-2) it is found that when the navigational figure I of input has Gradient Features, filtering output q also has similar gradient
Feature, therefore such filtering has guarantor's local edge.
To acquire filtering parameter a (x, y) and b (x, y), linear regression method is used to make the output valve of fitting function and true
Error between real value is minimum, even if formula (1-3) reaches minimum:
Wherein, P is image to be filtered, and ε is in order to avoid a (x, y) changes excessive regularization parameter.It is also to adjust filter
The important parameter of wave device filter effect.To ask error minimum, ask partial derivative that following equation group can be obtained simultaneously on above formula both sides:
Solving above-mentioned equation group can be obtained the filtering parameter a (x, y) as shown in formula (1-5), the value of b (x, y):
Wherein, μ (x, y) andMean value and variance of the navigational figure I in window N (x, y) are respectively indicated, | N (x, y) |
Indicate the pixel quantity in window N (x, y).It is the mean value of image to be filtered in the window.Calculating each window
Linear coefficient when, a pixel can include that is, each pixel is as described by multiple linear functions by multiple windows.Therefore,
, only need to be average by all linear function values comprising the point when specifically seeking the output valve of certain point, filtering output calculates
Formula is as follows:
Wherein,
When guiding filtering is used as holding edge filter device, i.e. I=P, if ε=0, it is clear that a=1, b=0 are formula
The solution of (1-3) minimum value, analysis know that filter at this time does not have any effect, will input intact output.If ε >
0, change small region (or monochromatic areas) in image pixel intensities, there is a to be similar to (or being equal to) 0, and b is similar to (or being equal to)A weighted mean filter is done;And changing big region, a is similar to 1, b and is similar to 0, the filtering to image
Effect is very weak, helps to maintain edge.And the effect of ε is exactly that is defined is that variation is big, what is that variation is small.In window size
In the case where constant, with the increase of ε, image filtering effect is more obvious.
Weight guiding filtering and the maximum difference of traditional guiding filtering are the adaptive determination of regularization parameter ε, pass
The guiding filtering of system is set to a constant, uses formula (1-7) in document, (1-8) calculates this parameter adaptive:
Wherein μ is constant, is taken as 1/128;ΓI(x, y) is guarantor's edge weights of navigational figure I, and (i, j) is with (x, y)
Centered on put pixel, any pixel in window M (x, y), | M (x, y) | indicate the pixel number in window M (x, y),
For variogram of the navigational figure in 3 × 3 matrix windows, ε0For constant, it is taken as (0.001 × 256)2。
2: the curvature of image is the important description of isophote morphological feature, can reflect image to a certain extent
Structure feature information, calculation formula is as shown in (1-9) formula:
Wherein, Ix(i,j),Iy(i,j),Ixx(i,j),Iyy(i, j) respectively represents image I on (i, j) point in the x-direction
Single order is led, the single order in the direction y is led, and the second order in the direction x is led, the second order in the direction y is led, and is calculated by formula (1-10):
3: the weight calculation formula (1-8) applied in WGIF also reflects the structure feature of image to a certain extent
The calculating of this weight is considered as a kind of transformation by information.
The information of reflecting edge and texture is data item in priority calculating function, but in some cases only according to number
It is difficult to efficiently differentiate out fringe region and texture region according to item.And calculation used by formula (1-11) then can be preferably
Solve the problems, such as this.
Function is calculated for the identification of picture structure and texture in order to improve priority, improves the standard of image repair effect
True property calculates function as priority using formula (1-11):
P (p)=C (p) [wD·D(p)+wK·|K(p)|+wΓ·ΓI(p)] (1-11)
Wherein, K (p) is curvature value of the image in p point, for the transformed value being calculated using formula (1-8);wD、wK、wΓ
For weight, and meet:
Calculate function according to new priority, in the less region of small curvature, image structure information, fill order grade compared with
It is low;And in deep camber, image structure information region abundant, priority level with higher, in this way, structure division information can
It is preferentially filled, thus obtained reparation result is more reasonable.
According to its empirical value of many experiments effect wD、wK、wΓIt is taken as 0.6,0.2,0.2 respectively
In the digital picture decomposable process based on gaussian pyramid, need to carry out gaussian filtering to image.And Gauss filters
Wave will cause the serious loss of image border and detailed information.
In order to overcome the problems, such as this, the present invention is combined adopt to image with the Image filter arithmetic for protecting edge effect
Sample decomposes, wherein having the Image filter arithmetic for protecting edge effect mainly to have bilateral filtering algorithm and navigational figure filtering algorithm
Deng by the analysis of front, the present invention uses weight guiding filtering WGIF, in order to obtain the general expression of WGIF, by its phase
It closes calculation formula to be deformed, central pixel point i and ω can be obtainediThe WGIF weighted value of any pixel j in neighborhoodIts calculation expression is as follows:
Wherein, k is i vertex neighborhood ωiWith j vertex neighborhood ωjAny pixel in intersection;|ωi|、|ωj| to be respectively neighbour
Domain ωi、ωjTotal pixel number;λWFor constant, its value is 1/128, μkFor the pixel average value where k point in window;∑k
For the standard deviation matrix of k point, WkFor the structural information value of k point, shown in corresponding calculation formula such as formula (4-14)~(4-18):
Wherein, [x], [y] are the r for indicating image, a certain channel in tri- channels g, b,For [x] of image I
Variogram of the channel components in 3 × 3 matrix windows, ε0For constant, value is (0.001 × 256)2。
Then according to above-mentioned correlation formula, obtained after carrying out multi-resolution decomposition using WGIF using the filter window of M × N size
To k scale tomographic image IkExpression formula are as follows:
Wherein m=[M/2], n=[N/2] are respectively represented no more than M/2, the maximum integer of N/2, WWGIF(u, v) is filtering
The corresponding WGIF filtering weighting value of pixel (u, v) in window.
In formula (1-20), Inew_k-1(i,j)R、Inew_k-1(i,j)G、Inew_k-1(i,j)BRespectively represent updated breakage
Image Inew_k-1In tri- component values of R, G, B of pixel (i, j), ηR、ηG、ηBThe respectively discrimination threshold in corresponding color channel,
The setting of its value follows formula (1-21).
Wherein ωmaskRepresent the image I of -1 scale layer of kthk-1Non- damaged area flag bit, | ωmask| in the region
Total pixel number, η0To compensate threshold value, it is taken as 3.
There are two " reparations " as a result, one is enterprising on kth scale layer image repair result basis in -1 scale layer of kth
Row up-sampling reconstruction obtainsThis is repaired indirectly as a result, another is then to use to be based on directly in -1 scale layer of kth
The algorithm of the priority of structure obtain it is direct reparation as a result, using similar in the two partially as common reparation as a result, from
And the damaged area range of original -1 scale layer of kth is reduced, the valid pixel in breakage image increases, and is subsequent
Reparation algorithm improve more reliable information.
In addition, in formula (1-21), seek two " reparation " results it is close whether threshold value when it is used be to ask flat
The thinking of mean value, i.e. calculating two images can embody in the corresponding absolute difference mean value in non-damaged area and directly repair and repair indirectly
Multiple difference degree introduces compensation threshold value η in formula0It is in order to avoid the judgment condition excessively harshness in formula (1-20) is led
Cause can not carry out effective information update.
The feasibility of this method is verified with specific test below, described below:
It is Intel i3-3110M, 2.4GHz that test result, which is this method in CPU, on the laptop for inside saving as 4G
Operation gained, operating system are Windows 7, and simulation software is 64 Matlab R2013a.
From figure 2 it can be seen that the Criminisi algorithm repaired using single scale image is for structure division
Repairing effect is not ideal enough, and repairing at yellow line has an obvious fracture, and the textured diffusion in road surface region the phenomenon that;Entropy
Algorithm, using traditional priority calculation, causes algorithm cannot good resolution technology and texture in repair process
The reparation in region repairs image and equally exists texture diffusion phenomena;And our rule the shortcomings that preferably overcoming above-mentioned algorithm,
Structure and texture obtain relatively reasonable natural repairing effect.
From figure 3, it can be seen that image-region to be repaired is near concrete column, there are grove and table top in the region being related to.
Criminisi algorithm repairing effect near texture information grove more abundant is poor, apparent fault-layer-phenomenon occurs, in platform
Face near border there are the diffusion of grove texture, reparation it is unreasonable.The effect that Entropy algorithm is repaired near grove is slightly better than
Criminisi algorithm, but the repairing effect near table top is still not ideal enough.And this method is to the structure repair near grove
Continuously naturally, gently reasonable to the reparation near table top, whole repairing effect is best.
In the image of Fig. 4 highway, the damaged area in complex pattern to be repaired includes apparent linear structure, passes through these three
Algorithm repairs the comparison of result as it can be seen that Criminisi algorithm fails are effectively repaired at structural break, and Entropy algorithm
Fail preferably to solve the problems, such as this.And our rule can preferably identify the structural information of image, and be drawn by weight
It leads filtering and carries out multi-resolution decomposition, ideal repairing effect is finally obtained by iteration reparation.
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Detailed process of the invention is as follows:
1) multiple dimensioned decomposition is carried out to image I to be repaired, i.e., since the 0th layer of smallest dimension layer, using based on weight
The algorithm of guiding filtering is filtered to each scale tomographic image and down-sampling, obtains a series of different scale tomographic image Ik;
2) since out to out layer, that is, top layer images, using structure-based prioritization functions calculation method, calculating is worked as
Preceding kth scale tomographic image IkThe priority of damaged area;
3) sequencing according to priority finds Optimum Matching using the similarity measurement criterion that corresponding matching criterior is constituted
Block carries out the image repair of current kth scale layer, obtains corresponding repairing effect figure Iipt_k;
4) current kth scale layer image repair result up-sampling reconstruction is carried out to obtainAs -1 scale of kth
Layer repair benchmark figure, and according to step 2), 3) repair -1 scale layer of kth image Ik-1It obtains first under this scale
Step repairs result Iinitial_ipt_k-1;
5) the preliminary reparation result I that will be generated in step 4)initial_ipt_k-1With benchmark figureIt compares, obtains
To new -1 scale layer breakage image I of kthnew_k-1;
6) according to step 2) and 3) to -1 scale layer breakage image I of kth newly obtainednew_k-1Image repair is carried out, is obtained
The final repairing effect figure I of -1 scale layer of kthipt_k-1;
7) it repeats step 4), 5) and 6) successively traverses each scale layer, until completing the reparation of smallest dimension layer, then repair
Terminate.
Claims (1)
1. a kind of multi-scale image restorative procedure based on structural information, including the following steps:
1) multiple dimensioned decomposition is carried out to image I to be repaired, i.e., since the 0th layer of smallest dimension layer, is guided using based on weight
The algorithm of filtering is filtered to each scale tomographic image and down-sampling, obtains a series of differences using the filter window of M × N size
Scale tomographic image Ik;Wherein, (i, j) is pixel coordinate,
M=[M/2], n=[N/2] are respectively represented no more than M/2, the maximum integer of N/2, WWGIF(u, v) is pixel in filter window
(u, v) corresponding WGIF filtering weighting value;
2) since out to out layer, that is, top layer images, the prioritization functions calculation method P (p) based on structural information=C is utilized
(p)·[wD·D(p)+wK·|K(p)|+wΓ·ΓI(p)] current kth scale tomographic image I, is calculatedkThe priority of damaged area;
Wherein C (p) is the confidence level of point p, and D (p) is the data item of point p, and K (p) is curvature, ΓI(p) factor is described for structural information,
wD、wK、wΓFor weight, and meet:
3) sequencing according to priority finds blocks and optimal matching blocks using the similarity measurement criterion that corresponding matching criterior is constituted,
The image repair for carrying out current kth scale layer obtains corresponding repairing effect figure Iipt_k;
4) current kth scale layer image repair result up-sampling reconstruction is carried out to obtainIt is repaired as -1 scale layer of kth
Multiple benchmark figure, and according to step 2), 3) repair -1 scale layer of kth image Ik-1Obtain tentatively repairing under this scale
Multiple junction fruit Iinitial_ipt_k-1;
5) the preliminary reparation result I that will be generated in step 4)initial_ipt_k-1With benchmark figureIt compares, obtains new
- 1 scale layer breakage image I of kthnew_k-1;
6) according to step 2) and 3) to -1 scale layer breakage image I of kth newly obtainednew_k-1Image repair is carried out, kth -1 is obtained
The final repairing effect figure I of scale layeript_k-1;
7) it repeats step 4), 5) and 6) successively traverses each scale layer, until completing the reparation of smallest dimension layer, then repair knot
Beam.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610963234.5A CN106530247B (en) | 2016-11-04 | 2016-11-04 | A kind of multi-scale image restorative procedure based on structural information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610963234.5A CN106530247B (en) | 2016-11-04 | 2016-11-04 | A kind of multi-scale image restorative procedure based on structural information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106530247A CN106530247A (en) | 2017-03-22 |
CN106530247B true CN106530247B (en) | 2019-07-09 |
Family
ID=58326329
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610963234.5A Active CN106530247B (en) | 2016-11-04 | 2016-11-04 | A kind of multi-scale image restorative procedure based on structural information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106530247B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107194897B (en) * | 2017-06-12 | 2019-07-23 | 大连海事大学 | A kind of precedence algorithm and image repair method based on architectural difference Yu marginal texture coefficient to be repaired |
CN107492073B (en) * | 2017-07-14 | 2020-08-14 | 天津大学 | Digital image restoration method based on weight guide filtering |
CN107507232B (en) * | 2017-07-14 | 2020-06-16 | 天津大学 | Stereo matching method based on multi-scale iteration |
CN107392943B (en) * | 2017-07-14 | 2020-06-16 | 天津大学 | Parallax refining method based on multi-scale weight guide filtering |
GB2569547B (en) * | 2017-12-19 | 2021-05-12 | Samsung Electronics Co Ltd | Reconstruction of original images from modified images |
CN109584178A (en) * | 2018-11-29 | 2019-04-05 | 腾讯科技(深圳)有限公司 | Image repair method, device and storage medium |
CN111325704B (en) * | 2020-02-07 | 2023-04-18 | 抖音视界有限公司 | Image restoration method and device, electronic equipment and computer-readable storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102117481A (en) * | 2011-03-17 | 2011-07-06 | 西安交通大学 | Automatic digital repair method of damaged images |
CN104574317A (en) * | 2015-01-19 | 2015-04-29 | 天津大学 | Image inpainting method based on isoluxline |
CN104680493A (en) * | 2015-03-12 | 2015-06-03 | 华东理工大学 | Digital image repairing method based on scale optimization |
CN105957027A (en) * | 2016-04-22 | 2016-09-21 | 西南石油大学 | MRF sample image restoring method based on required directional structural feature statistics |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101429365B1 (en) * | 2012-12-26 | 2014-08-13 | 중앙대학교 산학협력단 | Method and apparatus for Exemplar-Based Inpainting in a Multi-Scaled Space Using Laplacian Pyramid |
-
2016
- 2016-11-04 CN CN201610963234.5A patent/CN106530247B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102117481A (en) * | 2011-03-17 | 2011-07-06 | 西安交通大学 | Automatic digital repair method of damaged images |
CN104574317A (en) * | 2015-01-19 | 2015-04-29 | 天津大学 | Image inpainting method based on isoluxline |
CN104680493A (en) * | 2015-03-12 | 2015-06-03 | 华东理工大学 | Digital image repairing method based on scale optimization |
CN105957027A (en) * | 2016-04-22 | 2016-09-21 | 西南石油大学 | MRF sample image restoring method based on required directional structural feature statistics |
Also Published As
Publication number | Publication date |
---|---|
CN106530247A (en) | 2017-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106530247B (en) | A kind of multi-scale image restorative procedure based on structural information | |
CN110543911B (en) | Weak supervision target segmentation method combined with classification task | |
CN106204503B (en) | Based on the image repair algorithm for improving confidence level renewal function and matching criterior | |
CN102592268B (en) | Method for segmenting foreground image | |
CN103258203B (en) | The center line of road extraction method of remote sensing image | |
CN103325095B (en) | A kind of openness image repair method of sample block of the bonding position factor | |
CN111242864B (en) | Finger vein image restoration method based on Gabor texture constraint | |
CN108053417A (en) | A kind of lung segmenting device of the 3DU-Net networks based on mixing coarse segmentation feature | |
CN102521870B (en) | Coloring reuse method for micro-polygon ray tracing | |
CN106295613A (en) | A kind of unmanned plane target localization method and system | |
Fan et al. | Multi-scale feature fusion: Learning better semantic segmentation for road pothole detection | |
CN110992366A (en) | Image semantic segmentation method and device and storage medium | |
CN108629809B (en) | Accurate and efficient stereo matching method | |
Cao et al. | Restoration of an ancient temple mural by a local search algorithm of an adaptive sample block | |
CN102609903B (en) | A kind of method of the movable contour model Iamge Segmentation based on marginal flow | |
CN109101932A (en) | The deep learning algorithm of multitask and proximity information fusion based on target detection | |
CN103839274B (en) | A kind of Extended target tracking based on geometric proportion relation | |
CN103761717A (en) | City water extraction method based on panchromatic remote sensing image | |
CN104376312B (en) | Face identification method based on bag of words compressed sensing feature extraction | |
CN112241676A (en) | Method for automatically identifying terrain sundries | |
Wang et al. | A knowledge-based method for road damage detection using high-resolution remote sensing image | |
CN109740554A (en) | A kind of road edge line recognition methods and system | |
CN109410233A (en) | A kind of accurate extracting method of high-definition picture road of edge feature constraint | |
CN106296606B (en) | A kind of classification rarefaction representation image repair method of edge fitting | |
Song et al. | Building extraction from high resolution color imagery based on edge flow driven active contour and JSEG |
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 |