CN109903233A - A kind of joint image recovery and matching process and system based on linear character - Google Patents
A kind of joint image recovery and matching process and system based on linear character Download PDFInfo
- Publication number
- CN109903233A CN109903233A CN201910023228.5A CN201910023228A CN109903233A CN 109903233 A CN109903233 A CN 109903233A CN 201910023228 A CN201910023228 A CN 201910023228A CN 109903233 A CN109903233 A CN 109903233A
- Authority
- CN
- China
- Prior art keywords
- realtime graphic
- matrix
- pixel
- clear
- linear character
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The joint image that the invention discloses a kind of based on linear character restores and matching process, this method constructs Linear Mapping matrix, it extracts dominant linear feature calculation and initializes clear realtime graphic and weighting sparse coefficient, largest component in the weighting sparse coefficient of clear realtime graphic is selected, the position of element in a reference image in respective pixel dictionary matrix is final matching results.The present invention also provides the systems for realizing the above method.The present invention extracts linear character, plays and mitigates the fuzzy bring interference of image, a large amount of useless information of removal, holding reconstructs the effect of the linear relationship of clear realtime graphic with sparse coefficient, technical effect with raising image restoration and matching accuracy, while calculation amount is reduced, improve real-time.
Description
Technical field
The invention belongs to technical field of image processing, multiple more particularly, to a kind of joint image based on linear character
Former and matching process and system.
Background technique
Images match is the core technology in vision navigation system, is played a key effect to navigation accuracy is improved.In vision
In navigation system, video camera constantly shoots realtime graphic, and realtime graphic and the reference picture of storage are carried out scene matching, can be with
Obtain accurate location information.
Image matching algorithm is generally divided into method and method pixel-based based on feature.Images match based on feature
Algorithm is divided into two steps, and the first step is to extract feature in realtime graphic and reference picture, and second step is compared based on two groups of features
Matching result is obtained compared with similitude.The main research work of such algorithm is to extract different features, such as point feature
Sift feature, SURF feature, SUSAN feature, Canny feature and orthogonal lean wave characteristic for line feature etc..Experiment shows
Image matching algorithm based on feature has certain image affine transformation invariance, but when realtime graphic and reference picture are deposited
In noise and fuzzy situation, such algorithm is it is difficult to extract to accurate corresponding feature vector, therefore matching result is undesirable.
Image matching algorithm pixel-based is to be compared on a reference with sliding window extraction different images block and realtime graphic,
Because directly use all pixels, to there is a situation where noise and obscure available better matching result.Such is calculated
The main research work of method is how accurately to calculate the similitude, such as selective related coefficient, cross-correlation coefficient etc. of pixel.
Recently, sparse expression is applied in images match by Sai Yang et al., and the real-time of algorithm can be improved in this method, has one
Fixed antinoise is blocked with anti-.However, above-mentioned image matching algorithm all assumes that realtime graphic, there is no ambiguities.In reality
In, in picture system with it is external due to, image is inevitable in formation, record, processing and transmission process
Fuzzy equal degenerate problems are had, this makes images match become an extremely challenging problem.
Paper " Joint Image Restoration and Matching Based on Distance-Weighted
Sparse Representation " discloses a kind of joint image recovery and matched method for fuzzy realtime graphic, greatly
Cause scheme is that image when correct ambiguous estimation core, after recovery can be by dictionary most sparse expression, while rarefaction representation coefficient
The position of target can be positioned.On the one hand, rarefaction representation priori can be with the solution space of the possible clear image of operative constraint；Another party
Face, preferably restored image will be helpful to rarefaction representation, obtain better sparse expression coefficient, to obtain better positioning accurate
Degree.It is obscured since realtime graphic exists, the program is based on image one-dimensional pixel vector and calculates weighting sparse coefficient, and there are big spirograms
As the noise information influence of fuzzy introducing, so that sparse coefficient calculates inaccuracy, final image is caused to restore and matching result standard
Exactness decline.
Summary of the invention
There is fuzzy image for realtime graphic the purpose of the present invention is to provide a kind of in defect in view of the prior art
Matching process, it is intended to which the noise information for solving to obscure introducing there are image in existing method leads sparse coefficient calculating inaccuracy
The problem for causing matching precision low.
A kind of joint image recovery and matching process based on linear character, comprising the following steps:
(1) input step:
Input reference picture I and fuzzy realtime graphic y；
(2) pixel dictionary matrix D construction step:
Image block identical with fuzzy realtime graphic y size is extracted in sliding in reference picture I, by the picture of single image block
Element is stretched as column vector, then the pixel column vector of all image blocks is arranged to make up pixel dictionary matrix D；
(3) linear character dictionary matrix DfConstruction step:
The mapping matrix T for characterizing its linear character is constructed based on pixel dictionary matrix D, and then obtains linear character dictionary square
Battle array Df；
(4) clear realtime graphic x step is initialized:
The pixel of fuzzy realtime graphic y is stretched as column vector v1, vector v 1 and mapping matrix T-phase are multiplied to linear character
Column vector v2；In conjunction with linear characteristic series vector v 2 and linear character dictionary matrix DfThe weighting for calculating fuzzy realtime graphic y is sparse
Coefficient；Sparse coefficient is multiplied to obtain with pixel dictionary matrix D initializes clear realtime graphic x；
(5) iterative restoration and matching step:
Fuzzy core is calculated based on clear realtime graphic x and fuzzy realtime graphic y；Clear realtime graphic is updated according to fuzzy core
x；The pixel of the clear realtime graphic x of update is stretched as column vector v3, vector v 3 and mapping matrix T-phase are multiplied to linear special
Levy column vector v4；In conjunction with linear characteristic series vector v 4 and linear character dictionary matrix DfThe weighting for calculating clear realtime graphic x is dilute
Sparse coefficient；It repeats step (5), until reaching predetermined the number of iterations；
(6) matching result exports step:
Select largest component in the weighting sparse coefficient of clear realtime graphic x, the member in respective pixel dictionary matrix D
Position of the element in reference picture I is final matching results.
Further, the linear character dictionary matrix DfThe specific implementation of construction step are as follows:
Calculate the covariance matrix M of column vector in pixel dictionary matrix D；Solve the characteristic value and feature of covariance matrix M
Vector；M characteristic value is screened, the sum of the m characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；Extract this m
The corresponding feature vector of a characteristic value forms mapping matrix T；D is obtained into linear character dictionary matrix D multiplied by Tf。
Further, the linear character dictionary matrix DfThe specific implementation of construction step are as follows:
Element in pixel dictionary matrix D is changed into two dimensional image, is denoted as Ai, i=1,2 ..., N, N is dictionary element
Number；According to formulaCalculate the total population scatter matrix G of imaget, calculate Gt's
Characteristic value and feature vector；
M characteristic value is screened, the sum of the m characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；It extracts
The corresponding feature vector of this m characteristic value forms mapping matrix T1, calculate Linear Mapping characteristic pattern Fi=Ai·T1, i=1,
2,…,N；
Calculate total population scatter matrixCalculate GfCharacteristic value and spy
Levy vector；N characteristic value is screened, the sum of the n characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；Extract this
The corresponding feature vector of n characteristic value forms mapping matrix T2, calculate final Linear Mapping characteristic pattern Di=T2 T·Fi, i=1,
2 ..., N, by DiIt changes into column vector and constitutes linear character dictionary matrix Df。
Further, the specific implementation of the clear realtime graphic x step of initialization are as follows:
Fuzzy realtime graphic y is drawn into pixel column vector v 1, vector v 1 and mapping square according to row major order by (4-1)
Battle array T-phase is multiplied to arrive column vector v2；
(4-2) calculates column vector v2 and dictionary matrix DfIn each column vector Euclidean distance w1, by Euclidean distance w1
Vector v 2 is calculated in dictionary matrix D as weighting coefficientfOn sparse expression, obtain sparse coefficient α 1；
(4-3) sparse coefficient α 1 is multiplied to obtain with pixel dictionary matrix D initializes clear realtime graphic x.
Further, the specific implementation that fuzzy core is calculated based on clear realtime graphic x and fuzzy realtime graphic y
Are as follows:
Fuzzy coreWherein,
F () is Fast Fourier Transform (FFT), F ()-1It is inverse fast Fourier transform,It is the complex conjugate of F (),
Coefficient gamma value 0.003~0.007, I are unit matrix,It is matrix dot product.
Further, the specific implementation that clear realtime graphic x is updated according to fuzzy core are as follows:
According to objective function:
An auxiliary variable u is introduced, then objective function are as follows:
Wherein, k is the fuzzy core of previous step estimation, and y is fuzzy realtime graphic, and D is pixel dictionary matrix, e1=[1 ,-
1],e2=[1, -1]T, η, τ and β are weight parameters；
It solves to obtain clear realtime graphic x with the minimization of object function.
Further, clear realtime graphic x is solved using the method for alternating minimization, is first fixed u in each iteration and is come
Optimize x, then fixed x to optimize u；
Fixed u optimizes x:
Optimize x by solving following objective function:
The solution of object above function is:
Fix x again to optimize u:
Solve u:
A kind of joint image recovery and matching system based on linear character, comprises the following modules:
Input module, for inputting reference picture I and fuzzy realtime graphic y；
Pixel dictionary matrix D constructs module, multiple big with fuzzy realtime graphic y for sliding extraction in reference picture I
The pixel of single image block is stretched as column vector, then the pixel column vector of all image blocks is arranged by small identical image block
Constitute pixel dictionary matrix D；
Linear character dictionary matrix DfModule is constructed, characterizes its linear character for constructing based on pixel dictionary matrix D
Mapping matrix T, and then obtain linear character dictionary matrix Df；
Clear realtime graphic x module is initialized, for the pixel of fuzzy realtime graphic y to be stretched as column vector v1, vector
V1 and mapping matrix T-phase are multiplied to linear character column vector v2；In conjunction with linear characteristic series vector v 2 and linear character dictionary matrix
DfCalculate the weighting sparse coefficient of fuzzy realtime graphic y；Sparse coefficient is multiplied to obtain with pixel dictionary matrix D initializes clear reality
When image x；
Iterative restoration and matching module, for calculating fuzzy core based on clear realtime graphic x and fuzzy realtime graphic y；According to
Clear realtime graphic x is updated according to fuzzy core；The pixel of the clear realtime graphic x of update is stretched as column vector v3,3 He of vector v
Mapping matrix T-phase is multiplied to arrive linear character column vector v4；In conjunction with linear characteristic series vector v 4 and linear character dictionary matrix DfMeter
Settle the weighting sparse coefficient of clear realtime graphic x；Iteration restores and matching module, until reaching predetermined the number of iterations；
Matching result output module, largest component in the weighting sparse coefficient for selecting clear realtime graphic x correspond to
Position of the element in reference picture I in pixel dictionary matrix D is final matching results.
Contemplated above technical scheme through the invention, compared with prior art, the invention has the following advantages:
It is restored the present invention is based on the joint image of linear character and matching process is extracted due to constructing Linear Mapping matrix
Dominant linear feature initializes clear realtime graphic and weighting sparse coefficient for calculating, plays and mitigates the fuzzy bring of image
Interference, a large amount of useless information of removal, holding reconstruct the effect of the linear relationship of clear realtime graphic with sparse coefficient, have and mention
Hi-vision restores and the technical effect of matching accuracy, while can reduce algorithm calculation amount, improves algorithm real-time.
According to a kind of better embodiment, the Linear Mapping matrix of structuring one-dimensional vector is analyzed, it is specific special due to screening
Value indicative extracts corresponding feature vector composition Linear Mapping matrix, plays screening main information, remove the work of interference information
With.
According to a kind of better embodiment, the Linear Mapping matrix of analysis construction two dimensional image, due to respectively to X-Y scheme
The row and column of picture constructs Linear Mapping matrix, and total scatter matrix is identical with two dimensional image dimension, plays the master for keeping image airspace
Feature is wanted, the difficulty for calculating characteristic value and characteristic root is reduced, there is the technical effect for improving algorithm accuracy and real-time.
Detailed description of the invention
Fig. 1 is that the joint image of linear character restores and matching process flow chart.
Fig. 2 is that the joint image of linear character restores and matching process schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
The term used first to the present invention below is explained and illustrated.
Image restoration: image restoration is important one of the field of Digital Image Processing, is widely used in aerospace, astronomy
The fields such as observation, medical image diagnosis.In picture system with it is external due to, image is in formation, record, processing and biography
Noise is unavoidably had during defeated and obscures equal degenerate problems, so needing to restore image.Image restoration attempts
The image degenerated is rebuild or restored using certain priori knowledge of degradation phenomena, therefore recovery technique is exactly degradation model
Change, and handled using opposite process, to restore original image out.General pattern degenerative process can be modeled as
One degenrate function and an additive noise term, handle a width input picture f (x, y), generate a width degraded image g
(x,y).Given g (x, y) and some knowledge and additive noise item η (x, y) about degenrate function H, the purpose of image restoration
It is the approximate evaluation obtained about original imageIf system H is linear, location invariance a process,
The degraded image provided in the spatial domain can be given by:
G (x, y)=h (x, y) * f (x, y)+η (x, y)
According to whether known fuzzy convolution kernel, image restoration is broadly divided into non-blind deconvolution image restoration algorithm and blind warp
Product Image Restoration Algorithm.
Images match: image matching algorithm is generally divided into method and method pixel-based based on feature.Based on feature
Image matching algorithm be divided into two steps, the first step is to extract feature in realtime graphic and reference picture, and second step is based on two groups
Feature is compared similitude and obtains matching result.The main research work of such algorithm is to extract different features, such as needle
To the sift feature, SURF feature, SUSAN feature of point feature, Canny feature and orthogonal lean wave characteristic for line feature etc..
Image matching algorithm pixel-based is to be compared on a reference with sliding window extraction different images block and realtime graphic,
Because directly use all pixels, to there is a situation where noise and obscure available better matching result.Such is calculated
The main research work of method is how accurately to calculate the similitude, such as selective related coefficient, cross-correlation coefficient etc. of pixel.
Sparse expression: sparse expression assumes that data can be based on dictionary, is expressed as the linear combination of a part of element.It is given
Data x ∈ Rm, dictionary D=[d1,d2,…,dn]∈Rm×n, m≤n, we can be obtained based on sparse expression:
Above-mentioned model obtains the given x expression most sparse based on dictionary D.To conventional nonopiate and non-excessively complete dictionary
D, above-mentioned model solution are np hard problems, so generally using L1L is mentioned in regularization0Regularization is solved.Based on Lagrange
Conjugate Search Algorithm can be converted to following formula and be solved:
Fig. 1 provides preferably a kind of embodiment of the invention, specifically includes the following steps:
(1) input step:
Input reference picture I and fuzzy realtime graphic y；
(2) pixel dictionary matrix D is constructed:
(2-1) extracts image block B, the image block B length and width of extraction according to particular step size using sliding window on reference picture I
It is identical with realtime graphic y；
The pixel of image block B is drawn into a pixel column vector v according to row major order by (2-2)；
The pixel Column vector groups that all image blocks are drawn by (2-3) are pixel dictionary matrix D at a matrix；
(3) linear character dictionary matrix D is constructedfStep:
The first embodiment:
The mapping matrix T for characterizing its linear character is constructed based on pixel dictionary matrix D, pixel dictionary matrix D is multiplied by mapping
Matrix T obtains linear character dictionary matrix Df, more specific better embodiment are as follows:
The covariance matrix M of (3-1) calculating pixel dictionary matrix D；
(3-2) solves the characteristic value of covariance matrix M using Eigenvalues Decomposition；
(3-3) chooses m characteristic value, the sum of this m characteristic value divided by all characteristic values and greater than 0.9, take out this m
The corresponding feature vector of a characteristic value forms mapping matrix T；
(3-4) pixel dictionary matrix D obtains linear character dictionary matrix D multiplied by mapping matrix Tf。
Second of embodiment:
Element in pixel dictionary matrix D is changed into two dimensional image by (3-1), is denoted as Ai, i=1,2 ..., N, N is dictionary
Element number；According to formulaCalculate the total population scatter matrix G of imaget, meter
Calculate GtCharacteristic value and feature vector；
(3-2) screens m characteristic value, and the sum of the m characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；
Extract the corresponding feature vector composition mapping matrix T of this m characteristic value1, calculate Linear Mapping characteristic pattern Fi=Ai·T1, i=1,
2,…,N；
(3-3) calculates total population scatter matrixCalculate GfFeature
Value and feature vector；N characteristic value is screened, the sum of the n characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；
Extract the corresponding feature vector composition mapping matrix T of this n characteristic value2, calculate final Linear Mapping characteristic pattern Di=T2 T·Fi,i
=1,2 ..., N, by DiIt changes into column vector and constitutes linear character dictionary matrix Df。
(4) clear realtime graphic x is initialized:
The pixel of fuzzy realtime graphic y is stretched as column vector v1, vector v 1 and mapping matrix T-phase are multiplied to linear character
Column vector v2；In conjunction with linear characteristic series vector v 2 and linear character dictionary matrix DfThe weighting for calculating fuzzy realtime graphic y is sparse
Coefficient；Sparse coefficient is multiplied to obtain with pixel dictionary matrix D initializes clear realtime graphic x, more specific better embodiment
Are as follows:
Fuzzy realtime graphic y is drawn into pixel column vector v 1, vector v 1 and mapping matrix according to row major order by (4-1)
The multiplied column vector v2 to dimensionality reduction of T-phase；
(4-2) calculates vector v 2 and dictionary matrix DfIn each column vector Euclidean distance w1, Euclidean distance w1 make
Vector v 2 is calculated in dictionary matrix D for weighting coefficientfOn sparse expression, obtain sparse coefficient α；
(4-3) selectes n sparse coefficient component, and the sum of this n sparse coefficient component is divided by all sparse coefficient components
Be greater than predetermined threshold (for example 0.95), by this n sparse coefficient component and dictionary matrix DfMiddle corresponding element is multiplied to obtain just
The clear realtime graphic x of beginningization.
(5) iterative restoration and matching step:
Set the number of iterations T；Fuzzy core is calculated based on clear realtime graphic x and fuzzy realtime graphic y；More according to fuzzy core
New clear realtime graphic x；The pixel of the clear realtime graphic x of update is stretched as column vector v3, vector v 3 and mapping matrix T-phase
It is multiplied to arrive linear character column vector v4；In conjunction with linear characteristic series vector v 4 and linear character dictionary matrix DfCalculate clear figure in real time
As the weighting sparse coefficient of x；It repeats step (5), until reaching predetermined the number of iterations.
Calculate fuzzy core k:
According to formulaFuzzy core k is calculated, wherein x is clear realtime graphic, and y is fuzzy
Realtime graphic, F () are Fast Fourier Transform (FFT), F ()-1It is inverse fast Fourier transform,Be F () plural number it is total
Yoke, it is unit matrix that γ, which can take 0.005, I, and ο is matrix dot product
Update clear realtime graphic x:
According to objective function
An auxiliary variable u is introduced, then objective function are as follows:
Wherein, k is the fuzzy core of previous step estimation, and y is fuzzy realtime graphic, and D is pixel dictionary matrix, e1=[1 ,-
1],e2=[1, -1]T, s=0.5, β=0.015, η=1, τ=1.4.We are solved using the method for alternating minimization, substantially
Thought is iterative solution, u is fixed in each iteration first to optimize x, then fixes x to optimize u.
Optimize x:
Fixed u, optimizes x by solving following objective function:
The solution of object above function is:
Optimize u:
Fixed x, solves u:
Weighting sparse coefficient is calculated, is included the following steps
It will estimate that clear realtime graphic x is drawn into column vector v3 according to row major order, vector v 3 and mapping matrix T-phase multiply
Obtain the column vector v4 of dimensionality reduction；
Calculate vector v 4 and dictionary matrix DfIn each column vector Euclidean distance w2, using Euclidean distance w2 as plus
Weight coefficient calculates vector v 2 in dictionary matrix DfOn sparse expression, obtain sparse coefficient α
(6) matching result exports step:
The position of element in a reference image is exactly final in largest component respective pixel dictionary matrix D in sparse coefficient α
Matching result.
Example:
In order to compare the difference of the present invention and other methods, tested as follows.The reference of given 3 600*600 sizes
Image takes point centered on 100 identical coordinates to extract the picture of 49*49 size as real-time at random on every image
Image adds Gaussian Blur to realtime graphic, and the standard deviation of Gaussian Blur is 1 to 5, and setting sliding window extracts the step of dictionary image
A length of 1, it is tested using distinct methods.Realtime graphic is matched the sum of the pixel difference between coordinate and actual coordinate to be denoted as
PD, statistics PD are less than the sample proportion of different specific thresholds.As the following table 1 be joint image based on Weighted distance restore and
With algorithm experimental as a result, table 2 is the joint image recovery of the invention based on linear character and matching result, it can be seen that work as height
When this fuzzy standard deviation is 1, the matching precision of two methods is identical.But as the standard deviation of Gaussian Blur is gradually increased,
Joint image based on Weighted distance restores and the matching precision of matching algorithm sharply declines, when Gaussian Blur standard deviation is 5,
The sample proportion of PD≤6 is 0.62.Although the joint image based on linear character restores and matching also has under certain precision
Drop, but matching precision is more much higher than the former, when Gaussian Blur standard deviation is 5, the sample proportion of PD≤6 is 0.9767.
| PD≤0 | PD≤1 | PD≤2 | PD≤3 | PD≤4 | PD≤5 | PD≤6 | |
| σ=1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| σ=2 | 0.9433 | 0.9867 | 0.9933 | 0.9933 | 0.9933 | 0.9933 | 0.9933 |
| σ=3 | 0.4300 | 0.8100 | 0.9133 | 0.9400 | 0.9433 | 0.9433 | 0.9433 |
| σ=4 | 0.1400 | 0.4600 | 0.6767 | 0.7667 | 0.8067 | 0.8133 | 0.8133 |
| σ=5 | 0.0300 | 0.1633 | 0.3467 | 0.4733 | 0.5567 | 0.6000 | 0.6200 |
Table 1 is restored based on the joint image of Weighted distance and matching process experimental result
| PD≤0 | PD≤1 | PD≤2 | PD≤3 | PD≤4 | PD≤5 | PD≤6 | |
| σ=1 | 0.9900 | 1 | 1 | 1 | 1 | 1 | 1 |
| σ=2 | 0.9100 | 0.9800 | 1 | 1 | 1 | 1 | 1 |
| σ=3 | 0.6267 | 0.9000 | 0.9900 | 1 | 1 | 1 | 1 |
| σ=4 | 0.2767 | 0.6867 | 0.9167 | 0.9800 | 1 | 1 | 1 |
| σ=5 | 0.1267 | 0.4233 | 0.7200 | 0.8767 | 0.9367 | 0.9567 | 0.9767 |
Table 2 is restored based on the joint image of linear character and matching experimental result
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (9)
1. a kind of joint image based on linear character restores and matching process, which comprises the following steps:
(1) input step:
Input reference picture I and fuzzy realtime graphic y；
(2) pixel dictionary matrix D construction step:
Image block identical with fuzzy realtime graphic y size is extracted in sliding in reference picture I, and the pixel of single image block is drawn
It stretches for column vector, then the pixel column vector of all image blocks is arranged to make up pixel dictionary matrix D；
(3) linear character dictionary matrix DfConstruction step:
The mapping matrix T for characterizing its linear character is constructed based on pixel dictionary matrix D, and then obtains linear character dictionary matrix Df；
(4) clear realtime graphic x step is initialized:
The pixel of fuzzy realtime graphic y is stretched as column vector v1, vector v 1 and mapping matrix T-phase it is multiplied to linear character arrange to
Measure v2；In conjunction with linear characteristic series vector v 2 and linear character dictionary matrix DfCalculate the weighting sparse coefficient of fuzzy realtime graphic y；
Sparse coefficient is multiplied to obtain with pixel dictionary matrix D initializes clear realtime graphic x；
(5) iterative restoration and matching step:
Fuzzy core is calculated based on clear realtime graphic x and fuzzy realtime graphic y；Clear realtime graphic x is updated according to fuzzy core；It will
The pixel of the clear realtime graphic x updated is stretched as column vector v3, vector v 3 and mapping matrix T-phase it is multiplied to linear character arrange to
Measure v4；In conjunction with linear characteristic series vector v 4 and linear character dictionary matrix DfCalculate the weighting sparse coefficient of clear realtime graphic x；
It repeats step (5), until reaching predetermined the number of iterations；
(6) matching result exports step:
Largest component in the weighting sparse coefficient of clear realtime graphic x is selected, the element in respective pixel dictionary matrix D is being joined
Examining the position in image I is final matching results.
2. the joint image according to claim 1 based on linear character restores and matching process, which is characterized in that described
Linear character dictionary matrix DfThe specific implementation of construction step are as follows:
Calculate the covariance matrix M of column vector in pixel dictionary matrix D；Solve the characteristic value and feature vector of covariance matrix M；
M characteristic value is screened, the sum of the m characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；Extract this m feature
It is worth corresponding feature vector composition mapping matrix T；D is obtained into linear character dictionary matrix D multiplied by Tf。
3. the joint image according to claim 1 based on linear character restores and matching process, which is characterized in that described
Linear character dictionary matrix DfThe specific implementation of construction step are as follows:
Element in pixel dictionary matrix D is changed into two dimensional image, is denoted as Ai, i=1,2 ..., N, N is dictionary element number；It presses
According to formulaCalculate the total population scatter matrix G of imaget, calculate GtCharacteristic value
And feature vector；
M characteristic value is screened, the sum of the m characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；Extract this m
The corresponding feature vector of characteristic value forms mapping matrix T1, calculate Linear Mapping characteristic pattern Fi=Ai·T1, i=1,2 ..., N；
Calculate total population scatter matrixCalculate GfCharacteristic value and feature to
Amount；N characteristic value is screened, the sum of the n characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；Extract this n
The corresponding feature vector of characteristic value forms mapping matrix T2, calculate final Linear Mapping characteristic pattern Di=T2 T·Fi, i=1,2 ...,
N, by DiIt changes into column vector and constitutes linear character dictionary matrix Df。
4. the joint image according to claim 1 based on linear character restores and matching process, which is characterized in that described
Initialize the specific implementation of clear realtime graphic x step are as follows:
Fuzzy realtime graphic y is drawn into pixel column vector v 1, vector v 1 and mapping matrix T-phase according to row major order by (4-1)
It is multiplied to arrive column vector v2；
(4-2) calculates column vector v2 and dictionary matrix DfIn each column vector Euclidean distance w1, using Euclidean distance w1 as plus
Weight coefficient calculates vector v 2 in dictionary matrix DfOn sparse expression, obtain sparse coefficient α 1；
(4-3) sparse coefficient α 1 is multiplied to obtain with pixel dictionary matrix D initializes clear realtime graphic x.
5. the joint image according to claim 1 based on linear character restores and matching process, which is characterized in that described
The specific implementation of fuzzy core is calculated based on clear realtime graphic x and fuzzy realtime graphic y are as follows:
Fuzzy coreWherein,
F () is Fast Fourier Transform (FFT), F ()-1It is inverse fast Fourier transform,It is the complex conjugate of F (), coefficient
γ value 0.003~0.007, I are unit matrix, and ο is matrix dot product.
6. the joint image according to claim 1 based on linear character restores and matching process, which is characterized in that described
The specific implementation of clear realtime graphic x is updated according to fuzzy core are as follows:
According to objective function:
An auxiliary variable u is introduced, then objective function are as follows:
Wherein, k is the fuzzy core of previous step estimation, and y is fuzzy realtime graphic, and D is pixel dictionary matrix, e1=[1, -1], e2=
[1,-1]T, η, τ and β are weight parameters；
It solves to obtain clear realtime graphic x with the minimization of object function.
7. the joint image according to claim 6 based on linear character restores and matching process, which is characterized in that use
The method of alternating minimization solves clear realtime graphic x, u is fixed in each iteration first to optimize x, then fixed x and optimize u；
Fixed u optimizes x:
Optimize x by solving following objective function:
The solution of object above function is:
Fix x again to optimize u:
Solve u:
8. a kind of joint image based on linear character restores and matching system, which is characterized in that comprise the following modules:
Input module, for inputting reference picture I and fuzzy realtime graphic y；
Pixel dictionary matrix D constructs module, in reference picture I sliding extract it is multiple with fuzzy realtime graphic y size phase
Same image block, is stretched as column vector for the pixel of single image block, then the pixel column vector of all image blocks is arranged to make up
Pixel dictionary matrix D；
Linear character dictionary matrix DfModule is constructed, for constructing the mapping square for characterizing its linear character based on pixel dictionary matrix D
Battle array T, and then obtain linear character dictionary matrix Df；
Clear realtime graphic x module is initialized, for the pixel of fuzzy realtime graphic y to be stretched as column vector v1,1 He of vector v
Mapping matrix T-phase is multiplied to arrive linear character column vector v2；In conjunction with linear characteristic series vector v 2 and linear character dictionary matrix DfMeter
Calculate the weighting sparse coefficient of fuzzy realtime graphic y；Sparse coefficient is multiplied to obtain the clear figure in real time of initialization with pixel dictionary matrix D
As x；
Iterative restoration and matching module, for calculating fuzzy core based on clear realtime graphic x and fuzzy realtime graphic y；According to mould
It pastes core and updates clear realtime graphic x；The pixel of the clear realtime graphic x of update is stretched as column vector v3, vector v 3 and mapping
Matrix T-phase is multiplied to arrive linear character column vector v4；In conjunction with linear characteristic series vector v 4 and linear character dictionary matrix DfIt calculates clear
The weighting sparse coefficient of clear realtime graphic x；Iteration restores and matching module, until reaching predetermined the number of iterations；
Matching result output module, largest component in the weighting sparse coefficient for selecting clear realtime graphic x, respective pixel
Position of the element in reference picture I in dictionary matrix D is final matching results.
9. the joint image according to claim 8 based on linear character restores and matching system, which is characterized in that described
Linear character dictionary matrix DfConstruct the mapping matrix T's for characterizing its linear character in construction step based on pixel dictionary matrix D
Specific implementation are as follows:
The covariance matrix M of column vector in pixel dictionary matrix D is calculated, the characteristic value and feature vector of covariance matrix M are solved；
M characteristic value is screened, the sum of the m characteristic value and the ratio of the sum of all characteristic values are greater than predetermined threshold；Extract this m feature
It is worth corresponding feature vector composition mapping matrix T.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910023228.5A CN109903233B (en) | 2019-01-10 | 2019-01-10 | Combined image restoration and matching method and system based on linear features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910023228.5A CN109903233B (en) | 2019-01-10 | 2019-01-10 | Combined image restoration and matching method and system based on linear features |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109903233A true CN109903233A (en) | 2019-06-18 |
| CN109903233B CN109903233B (en) | 2021-08-03 |
Family
ID=66943626
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910023228.5A Expired - Fee Related CN109903233B (en) | 2019-01-10 | 2019-01-10 | Combined image restoration and matching method and system based on linear features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109903233B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103337058A (en) * | 2013-07-05 | 2013-10-02 | 西北工业大学 | Image blind restoration method based on blurred noise image pair joint optimization |
| US20140321738A1 (en) * | 2011-12-12 | 2014-10-30 | Nec Corporation | Dictionary creation device, image processing device, image processing system, dictionary creation method, image processing method, and program |
| CN104751420A (en) * | 2015-03-06 | 2015-07-01 | 湖南大学 | Sparse representation and multi-objective optimization based blind restoration method |
| JP2017034388A (en) * | 2015-07-30 | 2017-02-09 | 日本電信電話株式会社 | Dictionary generation method, dictionary generation apparatus, image processing method, image processing apparatus, and image processing program |
| CN107506718A (en) * | 2017-08-17 | 2017-12-22 | 南昌航空大学 | Facial expression recognizing method based on MB 2DPCA features |
| CN107563978A (en) * | 2017-08-31 | 2018-01-09 | 苏州科达科技股份有限公司 | Face deblurring method and device |
| CN108520497A (en) * | 2018-03-15 | 2018-09-11 | 华中科技大学 | Image restoration based on distance weighted sparse expression priori with match integral method |
-
2019
- 2019-01-10 CN CN201910023228.5A patent/CN109903233B/en not_active Expired - Fee Related
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140321738A1 (en) * | 2011-12-12 | 2014-10-30 | Nec Corporation | Dictionary creation device, image processing device, image processing system, dictionary creation method, image processing method, and program |
| CN103337058A (en) * | 2013-07-05 | 2013-10-02 | 西北工业大学 | Image blind restoration method based on blurred noise image pair joint optimization |
| CN104751420A (en) * | 2015-03-06 | 2015-07-01 | 湖南大学 | Sparse representation and multi-objective optimization based blind restoration method |
| JP2017034388A (en) * | 2015-07-30 | 2017-02-09 | 日本電信電話株式会社 | Dictionary generation method, dictionary generation apparatus, image processing method, image processing apparatus, and image processing program |
| CN107506718A (en) * | 2017-08-17 | 2017-12-22 | 南昌航空大学 | Facial expression recognizing method based on MB 2DPCA features |
| CN107563978A (en) * | 2017-08-31 | 2018-01-09 | 苏州科达科技股份有限公司 | Face deblurring method and device |
| CN108520497A (en) * | 2018-03-15 | 2018-09-11 | 华中科技大学 | Image restoration based on distance weighted sparse expression priori with match integral method |
Non-Patent Citations (7)
| Title |
|---|
| DAOQIANG ZHANG等: "(2D)2PCA: Two-directional two-dimensional PCA for efficient face representation and recognition", 《NEUROCOMPUTING》 * |
| HAICHAO ZHANG等: "Close the loop Joint blind image restoration and recognition with sparse representation prior", 《2011 INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| SHANSHAN WANG等: "Fenchel Duality Based Dictionary Learning for Restoration of Noisy Images", 《IEEE TRANSACTIONS ON IMAGE PROCESSING 》 * |
| YUANJIE SHAO等: "Joint Image Restoration and Matching Based on Distance-Weighted Sparse Representation", 《2018 24TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR)》 * |
| 刘文: "基于样例学习稀疏表示的非局部图像修复算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑（月刊）》 * |
| 廖海斌等: "基于高斯加权和因子分析字典学习的人脸姿态估计", 《扬州大学学报（自然科学版）》 * |
| 赵华波: "基于分块加权(2D)2PCA的单样本人脸识别", 《中国优秀硕士学位论文全文数据库 信息科技辑（月刊）》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109903233B (en) | 2021-08-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guo et al. | An efficient SVD-based method for image denoising | |
| Köhler et al. | Robust multiframe super-resolution employing iteratively re-weighted minimization | |
| Chan et al. | An augmented Lagrangian method for total variation video restoration | |
| Zhu et al. | Removing atmospheric turbulence via space-invariant deconvolution | |
| CN106952288B (en) | Based on convolution feature and global search detect it is long when block robust tracking method | |
| CN106228544B (en) | A kind of conspicuousness detection method propagated based on rarefaction representation and label | |
| Li et al. | FilterNet: Adaptive information filtering network for accurate and fast image super-resolution | |
| CN108305230A (en) | A kind of blurred picture integrated conduct method and system | |
| CN107680116B (en) | Method for monitoring moving target in video image | |
| Ren et al. | Partial deconvolution with inaccurate blur kernel | |
| CN109671029A (en) | Image denoising algorithm based on gamma norm minimum | |
| CN111160229B (en) | SSD network-based video target detection method and device | |
| Zhang et al. | Image deblurring using tri-segment intensity prior | |
| CN113570658A (en) | Monocular video depth estimation method based on depth convolutional network | |
| CN106683043B (en) | Parallel image splicing method and device of multi-channel optical detection system | |
| Hua et al. | Removing atmospheric turbulence effects via geometric distortion and blur representation | |
| Apdilah et al. | A study of Frei-Chen approach for edge detection | |
| CN113379647B (en) | Multi-feature image restoration method for optimizing PSF estimation | |
| CN112435211A (en) | Method for describing and matching dense contour feature points in endoscope image sequence | |
| Yang et al. | Deep convolutional grid warping network for joint depth map upsampling | |
| Solanki et al. | An efficient satellite image super resolution technique for shift-variant images using improved new edge directed interpolation | |
| CN113487491B (en) | Image restoration method based on sparsity and non-local mean self-similarity | |
| CN109903233A (en) | A kind of joint image recovery and matching process and system based on linear character | |
| CN111899284B (en) | Planar target tracking method based on parameterized ESM network | |
| CN114187198A (en) | Image strip removing method based on fuzzy matrix |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210803 Termination date: 20220110 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |