CN106023102B - A kind of image repair method based on Multi-scale model block - Google Patents
A kind of image repair method based on Multi-scale model block Download PDFInfo
- Publication number
- CN106023102B CN106023102B CN201610322871.4A CN201610322871A CN106023102B CN 106023102 B CN106023102 B CN 106023102B CN 201610322871 A CN201610322871 A CN 201610322871A CN 106023102 B CN106023102 B CN 106023102B
- Authority
- CN
- China
- Prior art keywords
- repaired
- block
- area
- edge pixels
- image
- 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 97
- 230000008439 repair process Effects 0.000 title claims abstract description 51
- 238000005259 measurement Methods 0.000 claims abstract description 32
- 230000006978 adaptation Effects 0.000 claims abstract description 4
- 238000012546 transfer Methods 0.000 claims description 9
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 230000000694 effects Effects 0.000 abstract description 19
- 230000001788 irregular Effects 0.000 abstract description 16
- 230000000007 visual effect Effects 0.000 abstract description 8
- 238000004364 calculation method Methods 0.000 abstract description 2
- 238000009792 diffusion process Methods 0.000 description 20
- 238000002474 experimental method Methods 0.000 description 9
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 238000005303 weighing Methods 0.000 description 2
- 241000287196 Asthenes Species 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 210000003462 vein Anatomy 0.000 description 1
Classifications
-
- G06T5/77—
-
- 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
Abstract
The invention discloses a kind of image repair methods based on Multi-scale model block, mainly solve the technical issues of prior art when repairing larger structure region to being easily distorted.The realization process includes: it is for area to be repaired edge pixels, passes through its inconsistency measurement and scale invariability metric calculation priority under different scale；To the highest area to be repaired edge pixels of current priority, block size to be repaired is determined according to its preferential level adaptation；The multiblock to be repaired is repaired using non local reconfiguration technique；Area to be repaired in traversing graph, repeats the above process, and terminates until repairing.The present invention uses different scale images block structure information, selects resulting structure block reparation, improves irregular texture and the judgement of random content priority, maintains image texture and structural integrity also inhibits artifact effect.The present invention obtains visual effect and the reparation of data target better image as a result, can be used for breakage image reparation and object removal.
Description
Technical field
The invention belongs to technical field of image processing, relate generally to image repair, specifically a kind of based on multiple dimensioned
The image repair method of block structure can be used for the recovery to natural image.
Background technique
Digital picture reparation is an important content in image restoration research, and the purpose is to according to the existing information of image
To restore to lose information automatically, can be used for removing the object of digital picture, and in text, work of fine arts or old photo
Scratch and lose the reparation of information etc..Therefore, digital picture reparation is in digital image processing field in occupation of very important
Status, become one of most basic technology in the field.But since the image of area to be repaired is unknown, digitized map
As having the technical problem much to be solved in repairing.
Existing restorative procedure can be roughly divided into two types, and one kind is the restorative procedure based on diffusion, and one kind is based on sample
The restorative procedure of example.
Restorative procedure based on diffusion is inherently a kind of restorative procedure based on partial differential equation, mainly there is base
Restorative procedure BSCB, the restorative procedure based on TV model, the reparation side based on Curvature-driven diffusion model in partial differential equation
Method etc., these methods have a repairing effect preferably repaired to small scale absent region, however it is larger for area to be repaired when, it is past
It is distorted toward will cause reparation.
Restorative procedure based on sample, thought are Future Opportunities of Texture Synthesis.On this basis, Criminisi et al. is proposed
A kind of image repair method based on sample priority.But the priority of Criminisi algorithm cannot be very to texture and structure
It is good to distinguish.In order to solve this problem, Sun Jian et al. proposes a kind of block diffusion image restorative procedure sparse based on block.But
It will appear the priority judgement to irregular grain region not based on the priority that the sparse block diffusion image of block repairs algorithm
Accurate problem.
The above-mentioned image repair method based on sample is all to make the circle circle of area to be repaired one by control confidence level item
Ecto-entad diffusion is repaired.When losing in image there are larger structure region, structure is caused to spread to two sides, so that structural area
It is distorted at domain.
The present invention proposes a kind of image repair method based on Multi-scale model block, designs new priority calculating side
Method, and according to the priority in the area to be repaired edge pixels with highest priority, adaptively adjust the to be repaired of the pixel
Multiblock size.The present invention measures priority by analyzing the structural information of image block under different scale, improves to image
The judgement of block structure information improves priority orders, and solves in Criminisi algorithm priority to texture and structure not
It can distinguish very well and priority is repaired in algorithm to the priority in irregular grain region based on the sparse block diffusion image of block
Judge the problem of inaccuracy.Meanwhile the present invention is adaptively adjusted the multiblock ruler to be repaired of highest priority pixel according to priority
It is very little, the directionality of well-formed is maintained, and inhibit the blurring effect often occurred in image repair to a certain extent.
Summary of the invention
It is an object of the invention to aiming at the problem that being distorted in above-mentioned prior art when repairing larger structure region,
A kind of image repair method based on Multi-scale model block is proposed, to pass through the structural information for analyzing image block on different scale
Priority is calculated, improves the accuracy to the judgement of texture region (including irregular grain) priority, while by preferential
The structural region that boundary is in intermediate is repaired, structural information is maintained, reduces the distortion to structure.
The present invention is a kind of image repair method based on Multi-scale model block, which is characterized in that is comprised the following steps that
(1) complex pattern I to be repaired is inputted, determines area to be repaired Ω and area to be repaired edge
(2) according to the Multi-scale model information of area to be repaired edge pixels p, all area to be repaired edge pixels are calculated
Priority,
(3) edge pixels of all area to be repaired edge pixels current highest priorities are set as q, according to the priority of q,
The block size f*f to be repaired of adaptive adjustment pixel q, and the multiblock to be repaired is set as the edge block ψ of highest priorityq；
(4) non local reconfiguration technique-Criminisi image repair method is applied, to the edge block ψ of highest priorityqInto
Row is repaired, and completes the reparation of an image block in complex pattern to be repaired, and update the area to be repaired edge
(5) step (2)-(4) are repeated, traverse area to be repaired Ω, Ω, which is all repaired, until area to be repaired finishes, extensive
It appears again the result images close with original image.
The present invention has used for reference the thought of the technical method based on sample, by the inconsistency degree for calculating block on different scale
Amount and scale invariability measurement are to measure the block structure information using area to be repaired edge pixels as block center.
The present invention has the advantage that compared with prior art
1. having used the structural information of image block on different scale in the present invention, the structure letter of image block has been measured well
Breath, preferentially picks out resulting structure block and is repaired, the priority for improving irregular texture region and random content region is sentenced
It is disconnected.
2. the present invention proposes the priority according to highest priority edge pixels, the block size to be repaired of the pixel is determined.
The advantages of not only being remained in this way based on sample image restorative procedure to texture repairing, and its structure maintain it is good consistent
Property, artifact effect is also inhibited to a certain extent.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention；
Fig. 2 is that image block divides diagram in step 2.1 of the invention；
Fig. 3 is first unbroken test image；
Fig. 4 is the first width containing damaged restored image to be repaired；
Fig. 5 is with existing two methods and the reparation result figure of the invention to Fig. 4；
Fig. 6 is second unbroken test image；
Fig. 7 is the second width containing damaged restored image to be repaired；
Fig. 8 is with existing two methods and the reparation result figure of the invention to Fig. 7；
Fig. 9 is the unbroken test image of third；
Figure 10 is third width containing damaged restored image to be repaired；
Figure 11 is the reparation result figure of existing two methods and the present invention to Figure 10.
Figure 12 is the detailed step explanatory diagram of the present invention；
Specific embodiment
Referring to the drawings, technical solutions and effects of the present invention is described in detail.
Embodiment 1:
The present invention is that a kind of image repair method based on Multi-scale model block is comprised the following steps that referring to Fig. 1
(1) it inputs complex pattern I to be repaired and determines area to be repaired Ω and area to be repaired edge referring to fig. 4
(2) it is directed to area to be repaired edge any pixel p,According to the multiple dimensioned of area to be repaired edge pixels p
Structural information calculates the priority of all area to be repaired edge pixels；
Wherein the Multi-scale model information of area to be repaired edge pixels p is made of two parts；Multi-scale model information
First part is inconsistency measurement of the area to be repaired edge pixels p on different scale, has measured area to be repaired side
Structural information in pixel p region.Specific practice is to calculate separately area to be repaired edge pixels p in block scale f*f
On and block scale f1*f1 on inconsistency measurement.Block scale f, f1 generally take 3~15 in image procossing, and this example is directed to
Fig. 4 sets f=7, f1=11.The second part of Multi-scale model information is the scale invariability of area to be repaired edge pixels p
Measurement, has measured the importance of its structural information in the region edge pixels p of area to be repaired；Specific practice is, according to
Ratio of the area to be repaired edge pixels p between the inconsistency measurement on block scale f*f and block scale f1*f1, meter
Calculate the scale invariability measurement of area to be repaired edge pixels p.
If the inconsistency measurement and scale invariability measurement of area to be repaired edge pixels p are higher, illustrate area to be repaired
Contain the Multi-scale model being more obvious in the domain region edge pixels p.Then the priority of area to be repaired edge pixels p is higher,
So the region area to be repaired edge pixels p, which more preferentially will be selected out, repairs.
(3) edge pixels of all area to be repaired edge pixels current highest priorities are set as q, according to the excellent of pixel q
The block size f*f to be repaired of first level adaptation adjustment pixel q, and the multiblock to be repaired is set as the edge block ψ of highest priorityq；
In image repair, the setting of the block size f size of multiblock to be repaired is very big to image repair influential effect.Block size
Be arranged it is excessive, then will lead to repair image in random content region when there is smoothing effect；And block size setting is too small, then can lead
Cause the phenomenon that structure dislocation occur when repairing structural region in image.Therefore the setting of the block size f size of multiblock to be repaired is very
It is crucial.The present invention is utilized according to the priority of the edge pixels q of current highest priority in all area to be repaired edge pixels
Its Multi-scale model information judges the content in the region where pixel q, and then adaptively adjusts the multiblock to be repaired of pixel q
Size f*f, and the multiblock to be repaired is set as the edge block ψ of highest priorityq.This example is directed to Fig. 4, if the region where pixel q
Content is structural region, then block size f=11；If the content in the region where pixel q is random content region, block size f
=5；If the content in the region where pixel q is Rule content region, block size f=7.
(4) non local reconfiguration technique-Criminisi image repair method is applied, to the edge block ψ of highest priorityqInto
Row is repaired, and completes the reparation of an image block in complex pattern to be repaired, and update the area to be repaired edge
(5) step (2)-step (4) are repeated, area to be repaired Ω is traversed, until area to be repaired Ω has all been repaired
Finish, recovers the result images close with original image.This example is repaired for the area to be repaired in Fig. 4, after repairing,
The result images close with original image are recovered, original image is referring to Fig. 3, and result figure is referring to Fig. 5 (c).
This example is in central location for Fig. 4 using the present invention and the area to be repaired being damaged is repaired, wherein
The content of area to be repaired mainly includes irregular texture information and structural information.The area to be repaired obtains after repairing
Result images are referring to Fig. 5 (c).Find out from Fig. 5 (c), the present invention studies the structural information of image block on different scale, using more
Mesostructure information is carried out the preferentially structural region picked out in area to be repaired and is repaired, improve irregular texture region and
The priority of structural region judges；Block size to be repaired is adaptively adjusted simultaneously, maintains good directionality in linear structure.
Embodiment 2:
With embodiment 1, step is believed in (2) according to Multi-scale model for a kind of image repair method based on Multi-scale model block
Breath, embodiment 2 calculate its priority for area to be repaired edge pixels all in Fig. 7.Pass through Multi-scale model information meter
Priority is calculated to comprise the following steps that
(2.a), in the image block of area to be repaired edge pixels, is to draw on f*f along direction in block scale to block center
Piecemeal, specific direction are evenly dividing block along direction j, and wherein direction j is made of the orthogonal direction group in plane coordinate system.This example
For Fig. 7, direction initialization j are as follows: in polar coordinate system in the plane, θ=0 °, 90 °, 45 °, -45 ° of this four directions, the four direction point
Other corresponding direction j=1,2,3,4；Referring to fig. 2.
For area to be repaired edge any pixel p,If using area to be repaired edge pixels p as block center
Image block is Ψp, and provide that the block size of the image block is f*f；This example is for Fig. 7, it is specified that block size f=7；By ΨpBe denoted as to
Image block of the restoring area edge pixels p on f*f block scale；Then, to ΨpAlong direction, j is evenly dividing image block, obtains one
Component masses pairWherein direction j is made of the orthogonal direction group in polar coordinate system in the plane.This example is directed to Fig. 7, setting side
To j are as follows: in polar coordinate system in the plane, θ=0 °, 90 °, 45 °, -45 ° of this four directions, the four direction respectively corresponds direction j=1,
2,3,4；Therefore, image block Ψ of the area to be repaired edge pixels p on f*f block scale is obtainedpFour component masses pair:Fig. 2 is shown in its division.
It is the inconsistency measurement on f*f, about direction j that (2.b), which calculates area to be repaired edge pixels p in block scale,.
According to ΨpPiecemeal pair on the j of directionArea to be repaired edge pixels p is calculated on f*f block scale,
Inconsistency about direction j measures difj(Ψp)；If P () is the operator for extracting unknown message in image block, thenJust
It is the operator for extracting Given information in image block；Simultaneously willA column vector is pulled into, which is denoted asAnd handle
First half content be denoted asThenLatter half content beWherein j=1,2,3,4；M=1,2.Then, to
Restoring area edge pixels p is in the inconsistency measurement dif that block scale is on f*f, about direction jj(Ψp) is defined as:
Wherein λ is equalizing coefficient, and this example is directed to Fig. 7, sets λ value 0.5；E () is the function for calculating pixel average.
For exampleFor statisticsThe function of the average value of middle all pixels value.
(2.c) is the inconsistency degree on f*f about all direction j in block scale according to area to be repaired edge pixels p
Amount, counting area to be repaired edge pixels p on the whole in block scale is the inconsistency measurement on f*f.
This example is directed to Fig. 7, and direction initialization j is made of 4 directions of polar coordinate system in the plane；According to area to be repaired edge picture
For plain p on f*f block scale, the inconsistency about this 4 directions measures difj(Ψp), wherein j=1,2,3,4；On the whole
It counts, inconsistency of the area to be repaired edge pixels p on f*f block scale measures Dif (Ψp)；
Wherein, Dif (Ψp) measure structural information of the area to be repaired edge pixels p on f*f block scale.
It is the inconsistency measurement on f1*f1 that (2.d), which calculates area to be repaired edge pixels p in block scale,
Referring to step (2.a)-step (2.c), if being Ψ by the image block at block center of area to be repaired edge pixels p
′p, and providing that the block size of the image block is f1*f1, this example is for Fig. 7, it is specified that block size f1=11；By Ψ 'pIt is denoted as to be repaired
Image block of the regional edge along pixel p on f1*f1 block scale；Then, according to image block Ψ 'p, calculate area to be repaired edge picture
Inconsistency of the plain p on f1*f1 block scale measures Dif (Ψ 'p)；
Wherein λ is equalizing coefficient, and this example is directed to Fig. 7, sets λ value 0.5；E () is the operator for calculating pixel average.
For exampleFor statisticsThe function of the average value of middle all pixels value.Dif(Ψ′p) measurement area to be repaired side
Along structural information of the pixel p on f1*f1 block scale.
(2.e) calculates the scale invariability measurement of area to be repaired edge pixels p
Inconsistency measurement according to area to be repaired edge pixels p on f*f block scale and on f1*f1 block scale: Dif
(Ψp) and Dif (Ψ 'p), the scale invariability for calculating area to be repaired edge pixels p measures Sca (p):
Wherein, scale invariability measurement Sca (p) has been measured its structure in the region edge pixels p of area to be repaired and has been believed
The importance of breath.
The priority of (2.f) calculating area to be repaired edge pixels p edge pixels
Dif (Ψ is measured using inconsistency of the area to be repaired edge pixels p on f*f block scalep) and Scale invariant
Property measurement Sca (p), calculate area to be repaired edge pixels p priority priority (p):
Wherein, T(ε)() is transfer function, is coordinate Multi-scale model block message and confidence information in priority flat
Weighing apparatus；By transfer function T(ε)() is defined as:Ifε is in transfer function
It adjustsThe setting value of value interval.This example is directed to Fig. 7, sets ε=6；WhereinFor unit vector.And C (p) is with area to be repaired
Image block confidence level item centered on the edge pixels p of domain, indicate image block centered on the edge pixels p of area to be repaired can
Letter degree；Confidence level CpIt (o) is the confidence level of pixel o in the image block；This example is directed to Fig. 7, to Cp(o) it is initialized: Cp
(o)=0；O ∈ Ω, Cp(o)=1；O ∈ (I- Ω) calculates image block Ψ using following formulapConfidence level item:
Wherein o is image block ΨpThe pixel of middle known pixel values.
In image repair method, the effect of priority is that the structure preferentially picked out in image is repaired, and is repaired
Effect is often related to the definition of structure to method.Criminisi method is using illumination line information come definition structure.It utilizes
Criminisi method repairs area to be repaired in Fig. 7, and result is found out referring to Fig. 8 (a) from Fig. 8 (a),
Criminisi method cannot keep linear structure well.Knot is utilized based on the sparse block diffusion image restorative procedure method of block
Structure degree of rarefication carrys out definition structure, repairs using based on the sparse block diffusion image restorative procedure of block to area to be repaired in Fig. 7
Multiple, result is referring to Fig. 8 (b).Find out from Fig. 8 (b), can not be protected well based on the sparse block diffusion image restorative procedure of block
Hold linear structure.And the present invention, from the property of the inconsistency of Multi-scale model block and scale invariability, utilization is multiple dimensioned
Structural information carrys out definition structure.This example using the present invention to be in central location in Fig. 7 and the area to be repaired that is damaged into
Row is repaired, and wherein the content of area to be repaired mainly contains various structures information.The area to be repaired obtains after repairing
Result images are referring to Fig. 8 (c).Find out that the present invention calculates priority by Multi-scale model information from Fig. 8 (c), preferentially picks out
Important structure efficiently solves the above two method problem.
Embodiment 3:
A kind of image repair method based on Multi-scale model block is with embodiment 1-2, and in step (3), embodiment 3 is for figure
All area to be repaired edge pixels in 10, if the area to be repaired edge pixels of its highest priority are q；Then according to pixel q
Priority, the block size f*f of the adaptive multiblock to be repaired for determining pixel q.
Specific practice includes: measuring Dif (Ψ according to the inconsistency in pixel q priorityq) (its block scale is f*f)
With scale consistent degree Sca (q), it is by pixel q points: structure edge pixels, random edge pixels or general edge pixels.
(3.a) is if Dif (Ψq) > Tresh2 and Sca (q) > Tresh1；Then pixel q belongs to structure edge pixels；If should
Multiblock to be repaired is the edge block ψ of highest priorityq, f:f=f1 in block size, wherein f1 > f；
(3.b) is if Dif (Ψq) < Tresh2 and Sca (q) < Tresh1；Then pixel q belongs to random edge pixels；If should
Multiblock to be repaired is the edge block ψ of highest priorityq, f:f=f2 in block size, wherein f2 < f；
(3.c) is other, then pixel q belongs to general edge pixels；If the multiblock to be repaired is the edge block ψ of highest priorityq,
F:f=f in its block size；
This example is directed to Figure 10, takes f=7, f1=11, f2=5, depending on the value visible image of Tresh1=0.75, Tresh2, knot
It is bigger that structure is more obvious then value；It sets Tresh2=max { Dif (p) }/2；
Block size to be repaired is very big on repairing effect influence in the restorative procedure based on sample, and common restorative procedure
In block size to be repaired be fixed.But it finds during the experiment, when the content for repairing image block to be repaired is linear structure
When, linear structure can be kept well if reparation block size is larger；And when the content for repairing image block to be repaired is more random,
The fuzzy artifact effect often occurred is avoided that if reparation block size is smaller.Therefore, the present invention is according to the priority of multiblock to be repaired,
Classified using the Multi-scale model information in priority to image block to be repaired, so that adaptively adjustment determines multiblock to be repaired
Block size.This example using the present invention to being in central location in Figure 10 and the area to be repaired that is damaged is repaired,
The content of middle area to be repaired mainly contains random content information and structural information.The area to be repaired obtains after repairing
To reparation the result is shown in Figure 11 (c).Find out from Figure 11 (c), the present invention using adaptive block size to be repaired to be repaired piece into
Row is repaired, and the good directionality of linear structure is not only maintained, and is also inhibited to a certain extent in image repair to random interior
Hold the blurring effect easily occurred when region is repaired, result figure has more natural visual effect after repairing it.
Embodiment 4:
With embodiment 1-3, this example provides entirety and in detail for a kind of image repair method based on Multi-scale model block
Example the present invention is further described.
Referring to Fig.1 2, of the invention the specific implementation steps are as follows:
Step 1, complex pattern I to be repaired is inputted, referring to fig. 4, Fig. 7, Figure 10.Above-mentioned three width figure region in an intermediate position is equal
It is damaged, needs to repair.Fig. 3, Fig. 6, image shown in Fig. 9 are the original image before Fig. 4, Fig. 7, Figure 10 are not damaged respectively.
It (is repaired in the case where there is original image referring to original image, if manually estimating original according to picture material to be repaired without original image
Figure.) it is directed to image graph 4 to be repaired, Fig. 7, Figure 10, it is necessary first to determine area to be repaired Ω and area to be repaired edge
Step 2, for area to be repaired edge any pixel p,If using area to be repaired edge pixels p as block
The image block at center is Ψp, wherein providing ΨpBlock size be f*f.By ΨpArea to be repaired edge pixels p is denoted as in f*f block
Image block on scale.Wherein f representative image block size, value 3~11 generally in image procossing；This example for Fig. 4, Fig. 7,
Figure 10 sets f value as 7.Pass through image block Ψ of the area to be repaired edge pixels p on f*f block scalep, calculate to be repaired
Regional edge measures Dif (Ψ along inconsistency of the pixel p on f*f block scalep)；
(2.1) it is directed to image block Ψ of the edge pixels p in area to be repaired on f*f block scalep, whereinBy Ψp
Along direction j, it is divided into two piecemealsWherein direction j is made of the orthogonal direction group in polar coordinate system in the plane.This example
For Fig. 4, Fig. 7, Figure 10, direction initialization j are as follows: in polar coordinate system in the plane, θ=0 °, 90 °, 45 °, -45 ° of this four directions, this four
A direction respectively corresponds direction j=1,2,3,4；Then four component masses pair are obtained: It is divided, and sees Fig. 2.
(2.2) according to ΨpThe piecemeal pair on the j of directionStatistical information, calculate area to be repaired edge pixels p
On f*f block scale, the inconsistency about direction j measures difj(Ψp).If P () is to extract unknown message in image block
Operator, thenIt is the operator for extracting Given information in image block.It willA column vector is pulled into, is denoted asIt will
First half content be denoted asThen latter half content isThenWithDivide equally jointlyInterior information,
Wherein j=1,2,3,4；M=1,2.Then, by area to be repaired edge pixels p on f*f block scale, about the different of direction j
Cause property measurement difj(Ψp) is defined as:
Wherein λ is equalizing coefficient, and this example is directed to Fig. 4, Fig. 7, Figure 10, sets λ value 0.5；E () is to calculate pixel to be averaged
The function of value.For exampleFor statisticsThe function of the average value of middle all pixels value.In the formula, previous item is anti-
Mirror ΨpPiecemeal pair on the j of directionBetween otherness, latter counts ΨpThe piecemeal pair on the j of directionInternal homogeneity information.
(2.3) according to area to be repaired edge pixels p on f*f block scale, the inconsistency about direction j measures difj
(Ψp), wherein j=1,2,3,4；Count the inconsistency measurement that can most represent pixel p on the whole on f*f block scale
Dif(Ψp)。
Wherein Dif (Ψp) structural information of the measurement area to be repaired edge pixels p on f*f block scale.
Step 3, for area to be repaired edge any pixel p,If using area to be repaired edge pixels p as block
The image block at center is Ψ 'p, wherein regulation Ψ 'pBlock size be f1*f1.By Ψ 'pArea to be repaired edge pixels p is denoted as to exist
Image block on f1*f1 block scale.This example is directed to Fig. 4, Fig. 7, Figure 10, sets f1 value as 11.Referring to step 2, calculate to be repaired
Multiple regional edge is the inconsistency measurement Dif (Ψ ' on f1*f1 scale in block along pixel pp)。
Wherein λ is equalizing coefficient, and this example is directed to Fig. 4, Fig. 7, Figure 10, sets λ value 0.5；E () is to calculate pixel to be averaged
The operator of value.For exampleFor statisticsThe function of the average value of middle all pixels value.Dif(Ψ′p) measurement it is to be repaired
Multiple structural information of the regional edge along pixel p on f1*f1 block scale.
Step 4, for area to be repaired edge any pixel p,According to area to be repaired edge pixels p in block
For the sum on f*f scale block be on f1*f1 scale inconsistency measurement be respectively Dif (Ψp) and Dif (Ψ 'p), slide ruler
Spend invariance measurement Sca (p).
Wherein, scale invariability measurement Sca (p) has been measured its structure in the region edge pixels p of area to be repaired and has been believed
The importance of breath.
Step 5, for area to be repaired edge any pixel p,According to area to be repaired edge pixels p in f*f
Inconsistency on scale measures Dif (Ψp) and scale invariability measurement Sca (p), calculate area to be repaired edge pixels p's
Priority priority (p).
Wherein, T(ε)() is transfer function, is coordinate Multi-scale model block message and confidence information in priority flat
Weighing apparatus；By transfer function T(ε)() is defined as:Ifε is to adjust in transfer function
SectionThe setting value of value interval.This example is directed to Fig. 4, Fig. 7, Figure 10, sets ε=6；WhereinFor unit vector；And C (p) be with
Image block Ψ centered on the edge pixels p of area to be repairedpConfidence level item, indicate with area to be repaired edge pixels o be
The image block Ψ of the heartpCredibility；Confidence level CpIt (o) is image block ΨpThe confidence level of middle pixel o；This example for Fig. 4, Fig. 7,
Figure 10, to Cp(o) it is initialized: Cp(o)=0；O ∈ Ω, Cp(o)=1；O ∈ (I- Ω) calculates image using following formula
Block ΨpConfidence level item:
Wherein o is image block ΨpThe pixel of middle known pixel values.
Step 6, if the area to be repaired edge pixels of all area to be repaired edge pixels current highest priorities are q,
It is adaptive to determine using pixel q as the block size size f*f of the multiblock to be repaired at block center according to the priority of pixel q.
Specific practice: Dif (Ψ is measured according to inconsistency of the pixel q on f*f scaleq) and consistency of scale measurement
Pixel q point is: structure edge pixels, random edge pixels or general edge pixels by Sca (q)；
(6.1) if Dif (Ψq) > Tresh2 and Sca (q) > Tresh1；Then pixel q belongs to structure edge pixels.If should
Multiblock to be repaired is the edge block ψ of highest priorityq, f:f=f1 in block size, (f1 > f).
(6.2) if Dif (Ψq) < Tresh2 and Sca (q) < Tresh1；Then pixel q belongs to random edge pixels.If should
Multiblock to be repaired is the edge block ψ of highest priorityq, f:f=f2 in block size, (f2 < f).
(6.3) other；Then pixel q belongs to general edge pixels.If the multiblock to be repaired is the edge block ψ of highest priorityq,
F:f=f in its block size.
This example is directed to Fig. 4, Fig. 7, Figure 10, takes f=7, f1=11, f2=5, the value view of Tresh1=0.75, Tresh2
As depending on, it is bigger that structure is more obvious then value；It sets Tresh2=max { Dif (p) }/2；
Step 7, using non local reconfiguration technique-Criminisi image repair method, to the edge block ψ of highest priorityq
It is repaired, completes the reparation of an image block in complex pattern to be repaired, and update the area to be repaired edgeIt is specific
Way includes:
(7.1) for the edge block ψ of highest priorityq, found in t × t contiguous range of pixel q by Euclidean distance
The image block most like with itThis example is directed to Fig. 4, Fig. 7, Figure 10, sets t=3*f.
Ns(q)={ ql: ql∈ N (q) and
Wherein, in t × t contiguous range of pixel q, for the images without unknown message all in the contiguous range
The set of its block center pixel is denoted as N by blocks(q), d () indicate Euclidean distance, N be pixel q t × t contiguous range in not
The number of image block containing unknown message, N (q) are the set of all pixels in t × t contiguous range of pixel q, while will be preferential
The highest edge block ψ of gradeqIn be located at area to be repaired pixel the value image block most like with itMiddle same position
The value of pixel covers；
Wherein u is the edge block ψ of highest priorityqIn be located at area to be repaired pixel.
(7.2) willMiddle corresponding ψqThe confidence level of area to be repaired same position pixel passes to ψqIn middle area to be repaired
Pixel.
WhereinWith the edge block ψ of highest priorityqMost like image blockThe confidence level of middle pixel u.
(7.3) to the edge block ψ of highest priorityqAfter reparation, the area to be repaired edge is updated
Step 8, step (2)-step (7) are repeated, area to be repaired Ω is traversed, until area to be repaired Ω is all repaired
It finishes, recovers and see close result images with original image.This example, by repairing, recovers and original image for Fig. 4, Fig. 7, Figure 10
As close result images, original image is respectively referring to Fig. 3, Fig. 6, Fig. 9；Its result figure is respectively referring to Fig. 5 (c), Fig. 8 (c), Figure 11
(c)。
This example is using the present invention for the area to be repaired point for being in central location in Fig. 4, Fig. 7, Figure 10 and being damaged
It is not repaired, after repairing, obtains result figure respectively referring to Fig. 5 (c), Fig. 8 (c), Figure 11 (c).From Fig. 5 (c), Fig. 8
(c), Figure 11 (c) finds out, the present invention calculates priority using Multi-scale model information, and preferentially being picked out by priority has
The image block of important feature is repaired, and the priority judgement of irregular texture region and structural region is not only improved, and
Correctly distinguish the difference of random content region and structural region；At the same time, the adaptively adjustment multiblock to be repaired in the present invention
Size mechanism not only maintains the good directionality of linear structure, but also inhibits repair random content area to a certain extent
The blurring effect that domain occurs often obtains showing good reparation result in visual effect.
Effect of the present invention is verified by following emulation experiment:
Embodiment 5:
A kind of image repair method based on Multi-scale model block is the same as embodiment 1-5.
Emulation experiment condition and method
Hardware platform are as follows: processor is Intel (R) Core (TM) i5-2450M CPU 2.50GHz, inside saves as 4.0G, firmly
Disk 320G, operating system are Microsoft windows sever 2007；
Software platform: MATLAB2014a；
Experimental method: being respectively existing Criminisi method and the block diffusion image restorative procedure sparse based on block
With method proposed by the invention.
Emulation content and result
Under above-mentioned experiment condition, one experiment of emulation is carried out
Emulation one central location and is damaged to be repaired using existing two methods and the present invention to being in Fig. 4
Region is repaired, and after repairing, result is referring to Fig. 5.Wherein Fig. 5 (a) is to use existing Criminisi method reparation
Result figure, Fig. 5 (b) are the result figure using the existing block diffusion image restorative procedure reparation sparse based on block, and Fig. 5 (c) is to make
The result figure repaired with the present invention.
Above-mentioned three kinds of methods are calculated separately to the Y-PSNR PSNR of the reparation result of area to be repaired in Fig. 4, are imitated
It is true that the results are shown in Table 1
Table 1 compares (unit: db) using the PSNR value that distinct methods repair result for Fig. 4
As seen from Table 1, the results show of existing two methods and the method for the present invention, Y-PSNR of the invention
PSNR is significantly increased.
Above-mentioned three kinds of methods are shown in Fig. 5 to the experimental result of Fig. 4 reparation respectively by this example respectively, wherein Fig. 5 (a), Fig. 5 (b)
It is compared with Fig. 5 (c) and non-breakage image Fig. 3.From visual effect, Fig. 5 (a) is that Criminisi method is to be repaired to Fig. 4
The result figure that multiple region is repaired finds out Fig. 5 (a) and non-breakage image Fig. 3 comparison, linear in central location in Fig. 5 (a)
Structure is interrupted by irregular grain, and irregular grain, which has been occupied, to be belonged to originally in the position of smooth region.Linear junction in Fig. 5 (a)
Structure be interrupted illustrate Criminisi method in Fig. 4 area to be repaired repair when, irregular grain area can not be distinguished well
Domain and structural region.Fig. 5 (b) is the result repaired based on the sparse block diffusion image restorative procedure of block to the area to be repaired Fig. 4
Fig. 5 (b) and non-breakage image Fig. 3 comparison is found out that the linear structure in Fig. 5 (b) in central location is destroyed by figure,
Irregular grain, which has been occupied, to be belonged to originally in the position of smooth region.Linear structure in Fig. 5 (b) is sparse based on block by explanation is destroyed
Block diffusion image restorative procedure in Fig. 4 area to be repaired repair when, can not distinguish well irregular grain region and
Structural region.Fig. 5 (c) is the result figure that the present invention repairs the area to be repaired Fig. 4, by Fig. 3 pairs of Fig. 5 (c) and non-breakage image
Than finding out, the linear structure region being damaged in the former area to be repaired of Fig. 5 (c) is restored well, irregular grain
Boundary between reason regional peace skating area domain is kept well.Linear structure in Fig. 5 (c), which is maintained, illustrates the present invention
When the image repair to be repaired containing irregular grain region and structural region this kind of to Fig. 4, the present invention can be distinguished well not
Regular veins region and structural region keep the structural information in complex pattern to be repaired.
Embodiment 6:
A kind of image repair method based on Multi-scale model block is with embodiment 1-5, and emulation experiment condition is the same as embodiment 5.
Emulation two utilizes existing two methods and of the invention, the area to be repaired for being in central location in Fig. 7 and being damaged
Domain is repaired, and after repairing, result is referring to Fig. 8.Wherein Fig. 8 (a) is the result repaired using Criminisi method
Figure, Fig. 8 (b) are using the result figure based on the sparse block diffusion image restorative procedure reparation of block, and Fig. 8 (c) is using the present invention
The result figure of reparation.
Above-mentioned three kinds of methods are calculated separately to the Y-PSNR PSNR of the reparation result of area to be repaired in Fig. 7, are imitated
It is true that the results are shown in Table 2
Table 2 compares (unit: db) using the PSNR value that distinct methods repair result to Fig. 7
As seen from Table 2, the results show of existing two methods and the method for the present invention, Y-PSNR of the invention
PSNR is significantly increased.
Above-mentioned three kinds of methods are shown in Fig. 8 to the experimental result of Fig. 7 reparation respectively by this example respectively, wherein Fig. 8 (a), Fig. 8 (b)
It is compared with Fig. 8 (c) and non-breakage image Fig. 6.From visual effect, Fig. 8 (a) is that Criminisi method is to be repaired to Fig. 7
Fig. 8 (a) and non-breakage image Fig. 6 comparison is found out that Fig. 8 (a) recovers multiple pyramid towers by the result figure that multiple region is repaired
Point.It, can not when extra pyramid pinnacle of a pagoda in Fig. 8 (a) illustrates that Criminisi method repairs area to be repaired in Fig. 7
The importance between various structures is distinguished well.Fig. 8 (b) is to be waited for based on the sparse block diffusion image restorative procedure of block Fig. 7
Fig. 8 (b) and non-breakage image Fig. 6 comparison is found out that the pyramid pinnacle of a pagoda in Fig. 8 (b) occurs by the result figure of restoring area reparation
Drift.The pyramid pinnacle of a pagoda to drift about in Fig. 8 (b) illustrates based on the sparse block diffusion image restorative procedure of block to be repaired in Fig. 7
When region is repaired, the importance distinguished between various structures can not be distinguished well.Fig. 8 (c) is that the present invention is to be repaired to Fig. 7
Fig. 8 (c) and non-breakage image Fig. 6 comparison is found out, is damaged in the former area to be repaired of Fig. 8 (c) by the result figure that region is repaired
Bad pyramid is restored well, and linear structure is maintained.Complete pyramid point illustrates the present invention in Fig. 8 (c)
When the complex pattern to be repaired containing various structures this kind of to Fig. 7, the present invention can distinguish the importance between various structures well, excellent
Resulting structure is first selected to be repaired.
Embodiment 7:
A kind of image repair method based on Multi-scale model block is with embodiment 1-6, and emulation experiment condition is the same as embodiment 6
Emulation three central location and is damaged to be repaired using existing two methods and the present invention to being in Figure 11
Multiple region is repaired, and after repairing, result is referring to Figure 11.Wherein Figure 11 (a) is repaired using Criminisi method
Result figure, Figure 11 (b) are using the result figure based on the sparse block diffusion image restorative procedure reparation of block, and Figure 11 (c) is to use
The result figure that the present invention repairs.
Since emulation three belongs to object removal image, non-reference picture is compared, therefore does not show its PSNR value.
Above-mentioned three kinds of methods are shown in Figure 11 to the experimental result of Figure 10 reparation respectively by this example respectively, wherein Figure 11 (a), Figure 11
(b) it is compared with Figure 11 (c) and non-breakage image Fig. 9.From visual effect, Figure 11 (a) is Criminisi method to figure
Figure 11 (a) and non-breakage image Fig. 9 comparison is found out that chad region occurs in Figure 11 (a) by the result figure that 10 areas to be repaired are repaired
(random content region) has corroded the edge (structural region) of former stepped area, so that step is broken.Fracture in Figure 11 (a)
Step illustrate Criminisi method in Figure 10 area to be repaired repair when, random content region can not be distinguished well
And structural region.Figure 11 (b) is the result repaired based on the sparse block diffusion image restorative procedure of block to the area to be repaired Figure 10
Figure 11 (b) and non-breakage image Fig. 9 comparison is found out that fracture also occurs in step in Figure 11 (b) by figure.It is broken in Figure 11 (b)
It, can not be regional very well when step illustrates to repair area to be repaired in Figure 10 based on the sparse block diffusion image restorative procedure of block
Divide random content region and structural region.Figure 11 (c) is the result figure that the present invention repairs the area to be repaired Figure 10, by Figure 11
(c) find out that the step of the Central Plains Figure 11 (c) area to be repaired is able to fully be restored, successfully with non-breakage image Fig. 9 comparison
Shelter in Fig. 9 is removed.Intact step illustrates that the present invention is this kind of to Figure 10 containing random content and knot in Figure 11 (c)
When the image repair to be repaired of structure, random content region and structural region can be distinguished well, its structural information is obtained
It keeps well.
Emulation experiment is summarized:
Since emulation three belongs to object removal image, non-reference picture is compared, therefore does not show its PSNR value.Then, on
The simulation experiment result is stated, it is as shown in table 3 by Data Summary.
Table 3 compares (unit: db) using the PSNR value that distinct methods repair result
From objective indicator Y-PSNR PSNR, as seen from Table 3, the method for the present invention is respectively to Fig. 4, Fig. 7 reparation
When, the Y-PSNR PSNR of result is improved than existing both of which.From subjective vision effect, the method for the present invention
It is that Fig. 7, Figure 10 are obtained when repairing as a result, referring to Fig. 5 (c) respectively to Fig. 4, Fig. 8 (c), Figure 11 (c), than existing two methods
It repairs as a result, repairing the view of result referring to Fig. 5 (a), Fig. 8 (a), Figure 11 (a) and Fig. 5 (b), Fig. 8 (b), Figure 11 (b) respectively
Feel that effect is more excellent than existing both of which.
Described in synthesis, above the experimental results showed that, the present invention no matter from visual effect or from data compare in other words without
By in objective indicator or subjective effect, preferable performance is all shown, while keeping effect structure, obtains height
The reparation result of quality.
In brief, a kind of image repair method based on Multi-scale model block disclosed by the invention, mainly solves existing
The problem of technology when repairing larger structure region to being easily distorted.Its realization process is: (1) according to the present invention propose based on
The priority calculation method of Multi-scale model block calculates priority to all area to be repaired edge pixels；(2) needed
It is true according to its preferential level adaptation to the area to be repaired edge pixels of current highest priority in restoring area edge pixels
The block size to be repaired of the fixed pixel, and the multiblock to be repaired is set as the edge block of highest priority.(3) non local reconstruct skill is applied
Art repairs the edge block of highest priority.(4) step (1)-step (3) are repeated, terminated until repairing.The present invention uses
Different scale block structure information selects resulting structure block reparation, improves irregular texture and the judgement of random content priority, protects
It has held image texture and structural integrity also inhibits artifact effect.By reparation of the invention, image is effectively restored, has obtained
Visual effect and the reparation of data target better image are as a result, can be used for breakage image reparation and object removal.
Claims (2)
1. a kind of image repair method based on Multi-scale model block, which is characterized in that comprise the following steps that
(1) complex pattern I to be repaired is inputted, determines area to be repaired Ω and area to be repaired edge
(2) it is directed to area to be repaired edge any pixel p,According to the Multi-scale model of area to be repaired edge pixels p
Information calculates the priority of all area to be repaired edge pixels, comprises the following steps that
(2.a) is directed to area to be repaired edge any pixel p,If using area to be repaired edge pixels p as block center
Image block is Ψp, and provide that the block size of the image block is f*f, by ΨpArea to be repaired edge pixels p is denoted as in f*f block ruler
Image block on degree；Then, to ΨpAlong direction, j is evenly dividing image block, obtains a component masses pairWherein direction j
It is made of the orthogonal direction group in plane coordinate system；Direction initialization j includes: in polar coordinate system in the plane, θ=0 °, and 90 °, 45 ° ,-
45 ° of this four directions, the four direction respectively correspond direction j=1, and 2,3,4；Therefore, area to be repaired edge pixels p can be obtained
Image block Ψ on f*f block scalepFour component masses pair:
(2.b) is according to ΨpPiecemeal pair on the j of directionArea to be repaired edge pixels p is calculated in f*f block scale
On, the inconsistency about direction j measures difj(Ψp)；If P () is the operator for extracting unknown message in image block, thenIt is the operator for extracting Given information in image block；Simultaneously willA column are pulled into, are denoted asIt willFirst half
Minute mark isThen latter half isWherein j=1,2,3,4；M=1,2；difj(Ψp) is defined as:
Wherein λ is equalizing coefficient, and E () is the function for calculating pixel average；For statisticsIn all pictures
The function of the average value of element value；
(2.c) on f*f block scale, measures dif about the inconsistency in all directions according to area to be repaired edge pixels pj
(Ψp), j=1,2,3,4；It counts on the whole, edge pixels p inconsistency on f*f block scale in area to be repaired measures Dif
(Ψp)；
Wherein, Dif (Ψp) measure structural information of the area to be repaired edge pixels p on f*f block scale；
(2.d) referring to step (2.a)-step (2.c), for area to be repaired edge any pixel p,If with to be repaired
Regional edge is Ψ ' along the image block that pixel p is block centerp, and provide that the block size of the image block is f1*f1, by Ψ 'pBe denoted as to
Image block of the restoring area edge pixels p on f1*f1 block scale；Then, according to image block Ψ 'p, calculate area to be repaired side
Dif (Ψ ' is measured along inconsistency of the pixel p on f1*f1 block scalep)；
Wherein λ is equalizing coefficient, and E () is the operator for calculating pixel average；For statisticsIn all pictures
The function of the average value of element value；Dif(Ψ′p) structural information of the measurement area to be repaired edge pixels p on f1*f1 block scale；
(2.e) measures Dif (Ψ according to inconsistency of the area to be repaired edge pixels p on different masses scalep) and Dif
(Ψ′p), the scale invariability for calculating area to be repaired edge pixels p measures Sca (p):
Wherein, scale invariability measurement Sca (p) has measured its structural information in the region edge pixels p of area to be repaired
Importance；
(2.f) measures Dif (Ψ using inconsistency of the edge pixels p in area to be repaired on f*f block scalep) and Scale invariant
Property measurement Sca (p), calculate area to be repaired edge pixels p priority priority (p):
Wherein, T(ε)() is transfer function, is the balance for coordinating Multi-scale model block message and confidence information in priority；
By transfer function T(ε)() is defined as:Ifε is to adjust in transfer functionThe setting value of value interval；WhereinFor unit vector；And C (p) is the image centered on the edge pixels p of area to be repaired
Block ΨpConfidence level item, indicate image block Ψ centered on the edge pixels p of area to be repairedpCredibility；Confidence level Cp
It (o) is image block ΨpThe confidence level of middle pixel o；To Cp(o) it is initialized: Cp(o)=0；O ∈ Ω, Cp(o)=1；o∈(I-
Ω), image block Ψ is calculated using following formulapConfidence level item:
Wherein o is image block ΨpThe pixel of middle known pixel values；
(3) it setting in all area to be repaired edge pixels, the area to be repaired edge pixels of current highest priority are q, according to
The block size f*f to be repaired of the preferential level adaptation adjustment pixel q of pixel q, and the multiblock to be repaired is set as the side of highest priority
Along block ψq；
(4) non local reconfiguration technique-Criminisi image repair method is applied, to the edge block ψ of highest priorityqIt is repaired
It is multiple, the reparation of an image block in complex pattern to be repaired is completed, and update the area to be repaired edge
(5) step (2)-step (4) are repeated, traverses area to be repaired Ω, Ω, which is all repaired, until area to be repaired finishes, extensive
It appears again the result images close with original image.
2. the image repair method according to claim 1 based on Multi-scale model block, which is characterized in that step (3), if
The area to be repaired edge pixels of all area to be repaired edge pixels current highest priorities are q, according to the preferential of pixel q
Grade, it is adaptive to determine using pixel q as the block size f*f of the multiblock to be repaired at block center；
Specific practice: Dif (Ψ is measured according to inconsistency of the pixel q on f*f scaleq) and consistency of scale measurement Sca (q),
It is by pixel q points: structure edge pixels, random edge pixels or general edge pixels；
(3.a) is if Dif (Ψq) > Tresh2 and Sca (q) > Tresh1；Then pixel q belongs to structure edge pixels；If this is to be repaired
Multiblock is the edge block ψ of highest priorityq, f:f=f1 in block size, wherein f1 > f；
(3.b) is if Dif (Ψq) < Tresh2 and Sca (q) < Tresh1；Then pixel q belongs to random edge pixels；If this is to be repaired
Multiblock is the edge block ψ of highest priorityq, f:f=f2 in block size, wherein f2 < f；
(3.c) is other, then pixel q belongs to general edge pixels；If the multiblock to be repaired is the edge block ψ of highest priorityq, block
F:f=f in size；
Wherein: f, f1, f2 are respectively the block size of different size of multiblock to be repaired, and relationship is f2 < f < f1；Tresh1,
Tresh2 is the edge block ψ in order to distinguish highest priorityqClassification and the threshold value that is arranged.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610322871.4A CN106023102B (en) | 2016-05-16 | 2016-05-16 | A kind of image repair method based on Multi-scale model block |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610322871.4A CN106023102B (en) | 2016-05-16 | 2016-05-16 | A kind of image repair method based on Multi-scale model block |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106023102A CN106023102A (en) | 2016-10-12 |
| CN106023102B true CN106023102B (en) | 2019-03-01 |
Family
ID=57098131
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610322871.4A Active CN106023102B (en) | 2016-05-16 | 2016-05-16 | A kind of image repair method based on Multi-scale model block |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106023102B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107507139B (en) * | 2017-07-28 | 2019-11-22 | 北京航空航天大学 | The dual sparse image repair method of sample based on Facet directional derivative feature |
| CN110675339A (en) * | 2019-09-16 | 2020-01-10 | 山东师范大学 | Image restoration method and system based on edge restoration and content restoration |
| CN111861943B (en) * | 2020-08-04 | 2024-02-20 | 北京惠朗时代科技有限公司 | Image restoration method for multi-scale pixel level depth optimization |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103578085A (en) * | 2013-10-18 | 2014-02-12 | 清华大学深圳研究生院 | Image empty region inpainting method based on variable blocks |
| CN103971338A (en) * | 2014-05-06 | 2014-08-06 | 清华大学深圳研究生院 | Variable-block image repair method based on saliency map |
-
2016
- 2016-05-16 CN CN201610322871.4A patent/CN106023102B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103578085A (en) * | 2013-10-18 | 2014-02-12 | 清华大学深圳研究生院 | Image empty region inpainting method based on variable blocks |
| CN103971338A (en) * | 2014-05-06 | 2014-08-06 | 清华大学深圳研究生院 | Variable-block image repair method based on saliency map |
Non-Patent Citations (2)
| Title |
|---|
| 动态尺度块匹配约束下的链式优化图像修复研究;许刚等;《电子学报》;20150331;第43卷(第3期);第529-535页 |
| 基于样本块结构特性的图像修复算法;范冠鹏;《中国优秀硕士学位论文全文数据库信息科技辑》;20131115(第11期);参见第2-3章 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106023102A (en) | 2016-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114418957B (en) | Global and local binary pattern image crack segmentation method based on robot vision | |
| JP4172941B2 (en) | Land parcel data creation method and apparatus | |
| CN106373123B (en) | Industrial component detection method of surface flaw based on k_tSL central cluster algorithm | |
| CN110570435B (en) | Method and device for carrying out damage segmentation on vehicle damage image | |
| CN106023102B (en) | A kind of image repair method based on Multi-scale model block | |
| CN109711268B (en) | Face image screening method and device | |
| CN108960135A (en) | Intensive Ship Target accurate detecting method based on High spatial resolution remote sensing | |
| KR20070023217A (en) | Method and apparatus for estimating orientation | |
| CN110827397B (en) | Texture fusion method for real-time three-dimensional reconstruction of RGB-D camera | |
| CN106373128B (en) | Method and system for accurately positioning lips | |
| CN102982524B (en) | Splicing method for corn ear order images | |
| CN108447022A (en) | Moving target joining method based on single fixing camera image sequence | |
| CN113810611B (en) | Data simulation method and device for event camera | |
| CN112102201A (en) | Image shadow reflection eliminating method and device, computer equipment and storage medium | |
| CN115471486A (en) | Switch interface integrity detection method | |
| CN107341824A (en) | A kind of comprehensive evaluation index generation method of image registration | |
| CN111415339A (en) | Image defect detection method for complex texture industrial product | |
| CN114674826A (en) | Visual detection method and detection system based on cloth | |
| WO2014205787A1 (en) | Vehicle detecting method based on hybrid image template | |
| CN112884795A (en) | Power transmission line inspection foreground and background segmentation method based on multi-feature significance fusion | |
| CN108682000B (en) | Pig body length and body width detection method based on Kinect video | |
| CN108447038A (en) | A kind of mesh denoising method based on non local full variation operator | |
| CN107657628A (en) | A kind of real-time color method for tracking target | |
| WO2021000302A1 (en) | Image dehazing method and system based on superpixel segmentation, and storage medium and electronic device | |
| CN107403429B (en) | Method for quickly and automatically acquiring parameters of periodic sequence image model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |