CN108537752A - Image processing method based on non local self-similarity and rarefaction representation and device - Google Patents
Image processing method based on non local self-similarity and rarefaction representation and device Download PDFInfo
- Publication number
- CN108537752A CN108537752A CN201810297729.8A CN201810297729A CN108537752A CN 108537752 A CN108537752 A CN 108537752A CN 201810297729 A CN201810297729 A CN 201810297729A CN 108537752 A CN108537752 A CN 108537752A
- Authority
- CN
- China
- Prior art keywords
- segment
- image
- sparse
- similarity
- preset
- 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
- 238000003672 processing method Methods 0.000 title claims abstract description 26
- 238000004364 calculation method Methods 0.000 claims description 20
- 239000012141 concentrate Substances 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 4
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 238000006467 substitution reaction Methods 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 11
- 230000006870 function Effects 0.000 description 20
- 238000000034 method Methods 0.000 description 17
- 238000010586 diagram Methods 0.000 description 10
- 238000001914 filtration Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241000406668 Loxodonta cyclotis Species 0.000 description 1
- 230000001351 cycling effect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
Classifications
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Abstract
The invention discloses a kind of image processing method and device based on non local self-similarity and rarefaction representation, the similar segment collection of each segment is calculated by introducing structural similarity amount SSIM, and by SSIM in the solution that the Euclidean distance of segment is combined the weighting coefficient for being applied to non local centralization sparse coding estimation, the weighting coefficient values of similar segment are effectively estimated, and then it calculates and obtains more accurate sparse coding value, operation is carried out with sparse coding estimated value to the preset dictionary of segment using iterative shrinkage algorithm simultaneously and obtains the sparse coefficient of each segment, sparse coefficient is substituted into preset reconstructed image function and carries out image reconstruction, obtain denoising image.Solves existing image sparse Denoising Algorithm sparse coefficient estimation inaccurate the technical issues of influencing denoising effect.
Description
Technical field
The present invention relates to technical field of image processing more particularly to a kind of based on non local self-similarity and rarefaction representation
Image processing method and device.
Background technology
With the development of image processing techniques, people require the identification of image also higher and higher.Pass through acquisition of taking pictures
Image, can all have a picture noise, the i.e. interference information unrelated with the true content of required image, and image processing techniques
One of which is exactly image denoising.
Newly interior is injected in the development that the non local self-similarity of segment is integrated as image processing techniques with sparse representation model
Appearance and vigor greatly push and promote the extensive use of Image Denoising Technology.The non local similitude of segment thinks segment in image
There is repeatability, i.e., any one segment can be found similar with it in image other positions on middle different spatial position
Segment will have the similar segment of structure be combined into a segment collection, however the measure of similitude such as Euclidean between segment
Distance Scaling Method, Cosin method, related coefficient measure etc. are measured using Euclidean distance, and segment is regarded as isolated pixel
Point ignores the structural similarity having between segment being handled.
To solve the problems, such as that the image processing effect caused by having ignored structural similarity is poor, a kind of current method is to draw
The index for having entered structural similarity estimates the similar segment of structure at (SSIM) measurement.Estimate in non local centralization sparse coding
In in order to obtain the original sparse coefficient of given segment rationally estimate traditional non local centralization sparse representation model first
The sparse coding of similar segment collection is weighted averagely to obtain estimated value by the similar segment collection for finding out segment, has ignored segment
Between the structural similarity that has so that in structure similar segment can it is excessive due to Euclidean distance and obtain smaller weighted value from
And cause the segment after weighted average that substantial deviation occurs with original tile so that the estimation of the original sparse coefficient of given segment
Inaccuracy influences image denoising effect.
Invention content
An embodiment of the present invention provides a kind of image processing method and dress based on non local self-similarity and rarefaction representation
It sets, is asked for solving the existing inaccurate technology for influencing denoising effect of image sparse Denoising Algorithm original sparse coefficient estimation
Topic.
A kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention, including：
S1：Acquisition waits for denoising image, waits for that denoising image is split to described, obtains several overlapped segments；
S2：The similarity that segment described in each two is calculated according to SSIM formula, obtains the similar segment of each segment
Collection, and calculate the Euclidean distance that the similar segment concentrates segment described in each two；
S3：The similarity of segment and the Euclidean distance described in each two is concentrated to add by preset according to the similar segment
Weight coefficient calculation formula calculates the similar segment and concentrates the weighting coefficient of each segment, and the weighting coefficient is substituted into
Preset sparse coding calculation formula calculates the sparse coding estimated value of each segment；
S4：It is calculated by iterative shrinkage according to the preset dictionary trained to the segment and the sparse coding estimated value
Method calculates the sparse coefficient of each segment, and the sparse coefficient is substituted into preset reconstructed image function and reconstructs to obtain first
Denoising image.
Preferably, step S4 is specifically included：
S41：The segment is clustered according to K mean algorithms, and the cluster is trained, obtains preset word
Allusion quotation；
S42：The segment is calculated by iterative shrinkage algorithm according to the preset dictionary and the sparse coding estimated value
Sparse coefficient, and the sparse coefficient is substituted into preset reconstructed image function and reconstructs to obtain denoising image.
Preferably, further include after step S4：
S5：The first denoising image and the Y-PSNR for waiting for denoising are calculated, the first denoising image is made
Denoising image, and return to step S1 are waited for for first；When the variation of the Y-PSNR tends towards stability, get and the peak
It is worth the optimal sparse coefficient of the corresponding denoising image of signal-to-noise ratio, and executes step S6；
S6：The excessively complete dictionary for calculating the preset dictionary, according to the excessively complete dictionary and the optimal sparse coefficient
Pass through reconstruction of function x=D αyReconstruct obtains denoising image, wherein αyFor optimal sparse coefficient, D was complete dictionary.
Preferably, the SSIM formula are：
Wherein,μy1And μy2For segment vector y1And y2In
The mean value of all pixels gray value, σy1And σy2For segment vector y1And y2The variance of grey scale pixel value, σy1y2For segment vector y1With
y2The covariance of grey scale pixel value, C1,C2And c3It is constant, and c3=c2/2。
Preferably, the preset weighting coefficient calculation formula is：
Wherein, xi=D αi, xi,q=D αi,q, D is the excessively complete dictionary of preset dictionary, αiFor the sparse of similar segment collection i
Degree, αi,qFor the degree of rarefication of q-th of segment in similar segment collection i, h is predefined scalar, and W is normaliztion constant, and n is each
The similar block number of image block.
Preferably, the preset sparse coding calculation formula is：
Wherein, ΩiFor similar segment collection, ωi,qFor q-th of segment weighting coefficient in similar segment collection i, αi,qFor similar diagram
The degree of rarefication of q-th of segment in block collection i.
Preferably, the preset reconstructed image function is：
Wherein, RkOperator is extracted for segment,For RkTransposition,It is sparse decomposition vector, φkIt is dictionary.
A kind of image processing apparatus of non local self-similarity and rarefaction representation, including：
Segment acquiring unit waits for denoising image for obtaining, waits for that denoising image is split to described, obtain several phases
The segment being mutually overlapped；
First computing unit, the similarity for calculating segment described in each two according to SSIM formula obtain each described
The similar segment collection of segment, and calculate the Euclidean distance that the similar segment concentrates segment described in each two；
Second computing unit, similarity and Euclidean distance for concentrating segment described in each two according to the similar segment
The weighting coefficient that the similar segment concentrates each segment is calculated by preset weighting coefficient calculation formula, and described will be added
Weight coefficient substitutes into the sparse coding estimated value that preset sparse coding calculation formula calculates each segment；
First image reconstruction unit estimates the preset dictionary that the segment is trained with the sparse coding for basis
Evaluation calculates the sparse coefficient of each segment by iterative shrinkage algorithm, and the sparse coefficient is substituted into preset reconstruct image
Transform reconstructs to obtain the first denoising image.
Preferably, image reconstruction unit specifically includes：
Training subelement is trained for being clustered to the segment according to K mean algorithms, and to the cluster,
Obtain preset dictionary；The segment is clustered according to K mean algorithms, and the cluster is trained, obtains preset word
Allusion quotation；
Image reconstruction subelement, based on according to the dictionary and the sparse coding estimated value by iterative shrinkage algorithm
Calculate the sparse coefficient of the segment.
Preferably, further include：
Third computing unit will be described for calculating the first denoising image and the Y-PSNR for waiting for denoising
First denoising image waits for denoising image, and return to step S1 as first；When the variation of the Y-PSNR tends towards stability,
The optimal sparse coefficient of denoising image corresponding with the Y-PSNR is got, and triggers the second image reconstruction unit；
Second image reconstruction unit, the excessively complete dictionary for calculating the preset dictionary, according to described excessively complete
Dictionary passes through reconstruction of function x=D α with the optimal sparse coefficientyReconstruct obtains denoising image, wherein αyFor optimal sparse system
Number, D was complete dictionary.
A kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention, including：S1：It obtains
It takes and waits for denoising image, wait for that denoising image is split to described, obtain several overlapped segments；S2：According to SSIM public affairs
Formula calculates the similarity of each two segment, obtains the similar segment collection of each segment, and calculates the similar segment and concentrate
The Euclidean distance of segment described in each two；S3：The similarity of segment described in each two and European is concentrated according to the similar segment
Distance calculates the weighting coefficient that the similar segment concentrates each segment by preset weighting coefficient calculation formula, and by institute
It states weighting coefficient and substitutes into the sparse coding estimated value that preset sparse coding calculation formula calculates each segment；S4：According to right
The preset dictionary that the segment is trained calculates each figure with the sparse coding estimated value by iterative shrinkage algorithm
The sparse coefficient of block, and the sparse coefficient is substituted into preset reconstructed image function and reconstructs to obtain the first denoising image.
A kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention is tied by introducing
The similar segment collection of each segment is calculated in structure similarity measure SSIM, and SSIM is combined application with the Euclidean distance of segment
In the solution for the weighting coefficient estimated to non local centralization sparse coding, the weighting coefficient values of similar segment are effectively estimated, into
And calculate and obtain more accurate sparse coding value, while using iterative shrinkage algorithm to the preset dictionary and sparse coding value of segment
It carries out operation and obtains the sparse coefficient of each segment, sparse coefficient, which is substituted into preset reconstructed image function, carries out image reconstruction, obtains
To denoising image.Solves the existing inaccurate skill for influencing denoising effect of image sparse Denoising Algorithm original sparse coefficient estimation
Art problem.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without having to pay creative labor, may be used also for those of ordinary skill in the art
To obtain other attached drawings according to these attached drawings.
Fig. 1 is one of a kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention
The flow diagram of embodiment；
Fig. 2 is a kind of the another of image processing method based on non local self-similarity and rarefaction representation provided by the invention
The flow diagram of a embodiment；
Fig. 3 is one of a kind of image processing apparatus based on non local self-similarity and rarefaction representation provided by the invention
The structural schematic diagram of embodiment；
Fig. 4 (a) is a kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention
One refers to segment schematic diagram；
Fig. 4 (b) is a kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention
One segment schematic diagram to be selected.
Fig. 4 (c) is a kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention
Another segment schematic diagram to be selected.
Specific implementation mode
An embodiment of the present invention provides a kind of collaborative filtering recommending method and devices, are pushed away for solving existing collaborative filtering
The result accuracy rate that the mode of recommending is recommended is not high enough, can not search information similar with the interest preference of new user according to scoring and go forward side by side
Row is recommended, the technical issues of being easy to cause user experience to reduce because of data sparsity problem.
In order to make the invention's purpose, features and advantages of the invention more obvious and easy to understand, below in conjunction with the present invention
Attached drawing in embodiment, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that disclosed below
Embodiment be only a part of the embodiment of the present invention, and not all embodiment.Based on the embodiments of the present invention, this field
All other embodiment that those of ordinary skill is obtained without making creative work, belongs to protection of the present invention
Range.
Referring to Fig. 1, Fig. 1 is a kind of image procossing based on non local self-similarity and rarefaction representation provided by the invention
The flow diagram of one embodiment of method.A kind of figure based on non local self-similarity and rarefaction representation provided by the invention
As one embodiment of processing method, including：
Step 101：Acquisition waits for denoising image, treats denoising image and is split, and obtains several overlapped segments.
It should be noted that usually any one of image other positions of segment in the images can all have it is similar
Segment, divide the image into several overlapped segments, that is, ensure that the similitude of segment；It can vividly describe
For：One size for waiting for denoising image is 3 × 3, by this image segmentation at 42 × 2 segments, then this 4 segments exist
It is overlapped.
Step 102：The similarity that each two segment is calculated according to SSIM formula, obtains the similar segment collection of each segment,
And calculate the Euclidean distance that similar segment concentrates each two segment.
It should be noted that as shown in figure 4, a) being one piece of regional area refers to segment, b) and c) be in region of search
The similar segment of search.Traditional similarity measurements flow function is calculated using Euclidean distance, can obtain segment a) and segment b) more phases
Seemingly；However people can visually find out a) and c) with structural similarity；Segment a) and segment c) are subjected to collaboration filtering
The architectural characteristic of segment can be preferably kept while filtering out segment noise；Segment a) and segment b) are subjected to collaboration filtering meeting
The original structure feature of denoising segment is destroyed, the texture and marginal information of image can be smoothed out when being finally spliced into original image.For
The structural similarity of segment is taken into account, therefore, introduces the similarity that structural similarity amount SSIM calculates each two segment.
Step 103：The similarity of each two segment and Euclidean distance is concentrated to pass through preset weighting coefficient according to similar segment
Calculation formula calculates the weighting coefficient that similar segment concentrates each segment, and weighting coefficient is substituted into preset sparse coding and calculates public affairs
Formula calculates the sparse coding estimated value of each segment.
It should be noted that the similarity of the similar segment of each two and Euclidean distance are combined, while considering segment
Between structural similarity and the Euclidean distance between segment, improve the method for solving of traditional segment weighting coefficient so that segment
Sparse estimation it is more accurate, avoiding similar segment in structure can be due to Euclidean distance is excessive and obtains weighting system
With artwork block substantial deviation occurs for the small segment so as to cause after weighted average of number so that the original sparse coefficient of given segment
Estimation is inaccurate, to influence the denoising effect of image.
Step 104：Pass through iterative shrinkage algorithm according to the preset dictionary and sparse coding estimated value trained to segment
The sparse coefficient of each segment is calculated, and sparse coefficient is substituted into preset reconstructed image function and reconstructs to obtain the first denoising image.
It should be noted that will wait for that denoising image is split after obtaining several overlapped segments, to segment
Preset dictionary corresponding with segment can be obtained by being trained, and the sparse coefficient for calculating each segment by iterative shrinkage algorithm has
Body formula is：Wherein, βiIt is to segment sparse coefficient αiReasonable estimation,
γ is regularization parameter, and p is constant 1 or 2.
A kind of image processing method based on non local self-similarity and rarefaction representation provided in an embodiment of the present invention, passes through
It introduces structural similarity amount SSIM and the similar segment collection of each segment is calculated, and the Euclidean distance of SSIM and segment are mutually tied
In the solution for closing the weighting coefficient for being applied to non local centralization sparse coding estimation, the weighting coefficient of similar segment is effectively estimated
Value, and then calculate and obtain more accurate sparse coding value, at the same using iterative shrinkage algorithm to the preset dictionary of segment with it is sparse
Encoded radio carries out operation and obtains the sparse coefficient of each segment, and sparse coefficient, which is substituted into preset reconstructed image function, carries out image weight
Structure obtains denoising image.Solves the inaccurate influence denoising effect of existing image sparse Denoising Algorithm original sparse coefficient estimation
The technical issues of fruit.
It is the one of a kind of image processing method based on non local self-similarity and rarefaction representation provided by the invention above
A embodiment is a kind of the another of image processing method based on non local self-similarity and rarefaction representation provided by the invention below
One embodiment.
Referring to Fig. 2, Fig. 2 is a kind of image procossing based on non local self-similarity and rarefaction representation provided by the invention
The flow diagram of another embodiment of method, it is provided by the invention a kind of based on non local self-similarity and rarefaction representation
Image processing method, including：
Step 201：Acquisition waits for denoising image, treats denoising image and is split, and obtains several overlapped segments.
It should be noted that usually any one of image other positions of segment in the images can all have it is similar
Segment, divide the image into several overlapped segments, that is, ensure that the similitude of segment；It can image point description
For：One size for waiting for denoising image is 3 × 3, by this image segmentation at 42 × 2 segments, then this 4 segments exist
It is overlapped.
Step 202：The similarity that each two segment is calculated according to SSIM formula, obtains the similar segment collection of each segment,
And calculate the Euclidean distance that similar segment concentrates each two segment.
It should be noted that as shown in figure 4, a) being one piece of regional area refers to segment, b) and c) be in region of search
The similar segment of search.Traditional similarity measurements flow function is calculated using Euclidean distance, can obtain segment a) and segment b) more phases
Seemingly；However people can visually find out a) and c) with structural similarity；Segment a) and segment c) are subjected to collaboration filtering
The architectural characteristic of segment can be preferably kept while filtering out segment noise；Segment a) and segment b) are subjected to collaboration filtering meeting
The original structure feature of denoising segment is destroyed, the texture and marginal information of image can be smoothed out when being finally spliced into original image.For
The structural similarity of segment is taken into account, therefore, introduces the similarity that structural similarity amount SSIM calculates each two segment.
Step 203：The similarity of each two segment and Euclidean distance is concentrated to pass through preset weighting coefficient according to similar segment
Calculation formula calculates the weighting coefficient that similar segment concentrates each segment, and weighting coefficient is substituted into preset sparse coding and calculates public affairs
Formula calculates the sparse coding estimated value of each segment.
It should be noted that the similarity of the similar segment of each two and Euclidean distance are combined, while considering segment
Between structural similarity and the Euclidean distance between segment, improve the method for solving of traditional segment weighting coefficient so that segment
Sparse estimation it is more accurate, avoiding similar segment in structure can be due to Euclidean distance is excessive and obtains weighting system
With artwork block substantial deviation occurs for the small segment so as to cause after weighted average of number so that the original sparse coefficient of given segment
Estimation is inaccurate, to influence the denoising effect of image.
Step 204：Segment is clustered according to K mean algorithms, and cluster is trained, obtains preset dictionary.
It should be noted that K mean algorithms are a kind of clustering algorithms, input is to cluster number k and comprising n data pair
The database of elephant, output are to meet k cluster of variance optimization criteria.To cluster centre according to Principal Component Analysis (PCA) into
Row training, can obtain corresponding preset dictionary.
Step 205：The sparse system of segment is calculated by iterative shrinkage algorithm according to preset dictionary and sparse coding estimated value
Number, and sparse coefficient is substituted into preset reconstructed image function and reconstructs to obtain denoising image.
It should be noted that the specific formula of the sparse coefficient for calculating each segment by iterative shrinkage algorithm is：
Wherein, βiIt is to segment sparse coefficient αiReasonable estimation, γ is
Regularization parameter, p are constants 1 or 2.
Step 206：It calculates the first denoising image and waits for the Y-PSNR of denoising, waited for using the first denoising image as first
Denoising image, and return to step 201；When the variation of the Y-PSNR tends towards stability, get and the peak value noise
Than the optimal sparse coefficient of corresponding denoising image, and execute step 207.
It should be noted that by the first denoising image that step 201 to step 205 obtains may not be it is optimal, because
This, is obtaining the first denoising image and after waiting for the Y-PSNR of denoising image, using the first denoising image as waiting for denoising figure
Picture, repetitive cycling step 201 to step 205, after iterative algorithm carries out continuous iteration to coefficient coefficient, Y-PSNR
It can constantly be optimized, can finally be tended towards stability, sparse coefficient when Y-PSNR being gone to tend towards stability is as optimal sparse
Coefficient, then step 207 is executed, it can learn, the denoising image denoising effect corresponding to optimal sparse coefficient is also optimal
's.
Step 207：The excessively complete dictionary for calculating preset dictionary, passes through according to excessively complete dictionary and the optimal sparse coefficient
Reconstruction of function x=D αyReconstruct obtains denoising image, wherein αyFor optimal sparse coefficient, D was complete dictionary.
It should be noted that if the atom in dictionary D be able to be turned into the theorem in Euclid space of n dimensions, dictionary D is complete
, if m>N, dictionary D are redundancies, while ensureing the theorem in Euclid space that can also be turned into n dimensions, then dictionary D was complete.It has crossed
The acquisition of standby dictionary is the prior art, herein without being described in detail.
Further, SSIM formula are：
Wherein,μy1And μy2For segment vector y1 and y2
The mean value of middle all pixels gray value, σy1And σy2For segment vector y1And y2The variance of grey scale pixel value, σy1y2For segment vector y1
And y2The covariance of grey scale pixel value, C1,C2And c3It is constant, and c3=c2/2。
Further, preset weighting coefficient calculation formula is：
Wherein, xi=D αi, xi,q=D αi,q, D is the excessively complete dictionary of preset dictionary, αiFor the sparse of similar segment collection i
Degree, αi,qFor the degree of rarefication of q-th of segment in similar segment collection i, h is predefined scalar, and W is normaliztion constant, and n is each
The similar block number of image block.
Further, preset sparse coding calculation formula is：
Wherein, ΩiFor similar segment collection, ωi,qFor q-th of segment weighting coefficient in similar segment collection i, αi,qFor similar diagram
The degree of rarefication of q-th of segment in block collection i.
Further, preset reconstructed image function is：
Wherein, RkOperator is extracted for segment, is RkTransposition,It is sparse decomposition vector, φkIt is dictionary.
It should be noted that table 1 be 13 width images are carried out with five kinds of algorithms under noise variance not of the same race it is hot-tempered and
The value of the Y-PSNR obtained, wherein Proposed are obtained according to the application algorithm, can prove to utilize the application's
The effect of the apparent algorithms more several than other of the denoising effect of algorithm is good.
Table 1
1 (Continued) of table
1 (Continued) of table
1 (Continued) of table
It is to a kind of another embodiment of the image processing method based on non local self-similarity and rarefaction representation above
Explanation, a kind of one embodiment of the image processing apparatus based on non local self-similarity and rarefaction representation is said below
It is bright.
Referring to Fig. 3, Fig. 3 is an a kind of reality of the image processing apparatus based on non local self-similarity and rarefaction representation
Apply example, a kind of image processing apparatus based on non local self-similarity and rarefaction representation provided by the invention, including：
Segment acquiring unit 301 waits for denoising image for obtaining, treats denoising image and be split, obtain several phases
The segment being mutually overlapped；
First computing unit 302, the similarity for calculating each two segment according to SSIM formula, obtains each segment
Similar segment collection, and calculate the Euclidean distance that similar segment concentrates each two segment；
Second computing unit 303, for concentrating the similarity of each two segment and Euclidean distance to pass through according to similar segment
Preset weighting coefficient calculation formula calculates the weighting coefficient that similar segment concentrates each segment, and weighting coefficient is substituted into preset dilute
Dredge the sparse coding estimated value that coding calculation formula calculates each segment；
First image reconstruction unit 304, for according to the preset dictionary trained to segment and sparse coding estimated value
The sparse coefficient of each segment is calculated by iterative shrinkage algorithm, and sparse coefficient is substituted into preset reconstructed image function and is reconstructed
To the first denoising image.
Further, image reconstruction unit 304 specifically includes：
Training subelement 3041 is trained for being clustered to segment according to K mean algorithms, and to cluster, obtains
Preset dictionary；Segment is clustered according to K mean algorithms, and cluster is trained, obtains preset dictionary；
Image reconstruction subelement 3042 is schemed for being calculated by iterative shrinkage algorithm according to dictionary and sparse coding estimated value
The sparse coefficient of block.
Further, further include：
Third computing unit 305, the Y-PSNR for calculating the first denoising image Yu waiting for denoising, described first is gone
Image of making an uproar waits for denoising image as first, and triggers image acquisition unit；When the variation of the Y-PSNR tends towards stability,
The optimal sparse coefficient of denoising image corresponding with the Y-PSNR is got, and triggers the second image reconstruction unit 306；
Second image reconstruction unit 306, the excessively complete dictionary for calculating preset dictionary, according to excessively complete dictionary with it is optimal
Sparse coefficient passes through reconstruction of function x=D αyReconstruct obtains denoising image, wherein αyFor optimal sparse coefficient, D was complete word
Allusion quotation.
The above, the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although with reference to before
Stating embodiment, invention is explained in detail, it will be understood by those of ordinary skill in the art that：It still can be to preceding
The technical solution recorded in each embodiment is stated to modify or equivalent replacement of some of the technical features；And these
Modification or replacement, the spirit and scope for various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution.
It is apparent to those skilled in the art that for convenience and simplicity of description, the device of foregoing description,
The specific work process of device and module, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In several embodiments provided herein, it should be understood that disclosed device, device and method can be with
It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the division of module,
Only a kind of division of logic function, formula that in actual implementation, there may be another division manner, such as multiple module or components can be with
In conjunction with or be desirably integrated into another device, or some features can be ignored or not executed.Another point, it is shown or discussed
Mutual coupling, direct-coupling or communication connection can be by some interfaces, the INDIRECT COUPLING of device or module or
Communication connection can be electrical, machinery or other forms.
The module illustrated as separating component may or may not be physically separated, and be shown as module
Component may or may not be physical module, you can be located at a place, or may be distributed over multiple networks
In module.Some or all of module therein can be selected according to the actual needs to achieve the purpose of the solution of this embodiment.
In addition, each function module in each embodiment of the present invention can be integrated in a processing module, it can also
That modules physically exist alone, can also two or more modules be integrated in a module.Above-mentioned integrated mould
The form that hardware had both may be used in block is realized, can also be realized in the form of software function module.
If integrated module is realized and when sold or used as an independent product in the form of software function module, can
To be stored in a computer read/write memory medium.Based on this understanding, technical scheme of the present invention substantially or
Say that all or part of the part that contributes to existing technology or the technical solution can embody in the form of software products
Out, which is stored in a storage medium, including some instructions are used so that a computer equipment
(can be personal computer, server or the network equipment etc.) executes all or part of each embodiment method of the present invention
Step.And storage medium above-mentioned includes：It is USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random
Access various Jie that can store program code such as memory (RAM, Random Access Memory), magnetic disc or CD
Matter.
Claims (10)
1. a kind of image processing method based on non local self-similarity and rarefaction representation, which is characterized in that including：
S1：Acquisition waits for denoising image, waits for that denoising image is split to described, obtains several overlapped segments；
S2：The similarity that segment described in each two is calculated according to SSIM formula, obtains the similar segment collection of each segment, and
Calculate the Euclidean distance that the similar segment concentrates segment described in each two；
S3：It is by preset weighting according to the similarity of segment and the Euclidean distance described in the similar segment concentration each two
Number calculation formula calculates the weighting coefficient that the similar segment concentrates each segment, and weighting coefficient substitution is preset
Sparse coding calculation formula calculates the sparse coding estimated value of each segment；
S4：Pass through iterative shrinkage algorithm meter according to the preset dictionary and the sparse coding estimated value trained to the segment
The sparse coefficient of each segment is calculated, and the sparse coefficient is substituted into preset reconstructed image function and reconstructs to obtain the first denoising
Image.
2. the image processing method of non local self-similarity and rarefaction representation according to claim 1, which is characterized in that step
Rapid S4 is specifically included：
S41：The segment is clustered according to K mean algorithms, and the cluster is trained, obtains preset dictionary；
S42：The dilute of the segment is calculated by iterative shrinkage algorithm according to the preset dictionary and the sparse coding estimated value
Sparse coefficient, and the sparse coefficient is substituted into preset reconstructed image function and reconstructs to obtain the first denoising image.
3. the image processing method of non local self-similarity and rarefaction representation according to claim 2, which is characterized in that step
Further include after rapid S4：
S5：The first denoising image and the Y-PSNR for waiting for denoising are calculated, using the first denoising image as
One waits for denoising image, and return to step S1；When the variation of the Y-PSNR tends towards stability, gets and believe with the peak value
It makes an uproar than the optimal sparse coefficient of corresponding denoising image, and executes step S6；
S6：The excessively complete dictionary for calculating the preset dictionary passes through according to the excessively complete dictionary with the optimal sparse coefficient
Reconstruction of function x=D αyReconstruct obtains denoising image, wherein αyFor optimal sparse coefficient, D was complete dictionary.
4. the image processing method of non local self-similarity and rarefaction representation according to claim 1, which is characterized in that institute
Stating SSIM formula is：
Wherein,μy1And μy2For segment vector y1And y2In own
The mean value of grey scale pixel value, σy1And σy2For segment vector y1And y2The variance of grey scale pixel value, σy1y2For segment vector y1And y2Picture
The covariance of plain gray value, C1,C2And c3It is constant, and c3=c2/2。
5. the image processing method of non local self-similarity and rarefaction representation according to claim 3, which is characterized in that institute
Stating preset weighting coefficient calculation formula is：
Wherein, xi=D αi, xi,q=D αi,q, D is the excessively complete dictionary of preset dictionary, αiFor the degree of rarefication of similar segment collection i, αi,q
For the degree of rarefication of q-th of segment in similar segment collection i, h is predefined scalar, and W is normaliztion constant, and n is each image block
Similar block number.
6. the image processing method of non local self-similarity and rarefaction representation according to claim 1, which is characterized in that institute
Stating preset sparse coding calculation formula is：
Wherein, ΩiFor similar segment collection, ωi,qFor q-th of segment weighting coefficient in similar segment collection i, αi,qFor similar segment collection i
In q-th of segment degree of rarefication.
7. the image processing method of non local self-similarity and rarefaction representation as claimed in any of claims 2 to 6,
It is characterized in that, the preset reconstructed image function is：
Wherein, RkOperator is extracted for segment,For RkTransposition,It is sparse decomposition vector, φkIt is dictionary.
8. a kind of image processing apparatus of non local self-similarity and rarefaction representation, which is characterized in that including：
Segment acquiring unit waits for denoising image for obtaining, waits for that denoising image is split to described, obtain several phase mutual respects
Folded segment；
First computing unit, the similarity for calculating segment described in each two according to SSIM formula, obtains each segment
Similar segment collection, and calculate the Euclidean distance that the similar segment concentrates segment described in each two；
Second computing unit, for concentrating the similarity of segment and Euclidean distance described in each two to pass through according to the similar segment
Preset weighting coefficient calculation formula calculates the weighting coefficient that the similar segment concentrates each segment, and is by the weighting
Number substitutes into the sparse coding estimated value that preset sparse coding calculation formula calculates each segment；
First image reconstruction unit, for according to the preset dictionary trained to the segment and the sparse coding estimated value
The sparse coefficient of each segment is calculated by iterative shrinkage algorithm, and the sparse coefficient is substituted into preset reconstructed image letter
Number reconstruct obtains the first denoising image.
9. the image processing apparatus of a kind of non local self-similarity and rarefaction representation according to claim 8, feature exist
In image reconstruction unit specifically includes：
Training subelement is trained for being clustered to the segment according to K mean algorithms, and to the cluster, obtains
Preset dictionary；The segment is clustered according to K mean algorithms, and the cluster is trained, obtains preset dictionary；
Image reconstruction subelement, for calculating institute by iterative shrinkage algorithm according to the dictionary and the sparse coding estimated value
State the sparse coefficient of segment.
10. the image processing apparatus of a kind of non local self-similarity and rarefaction representation according to claim 9, feature exist
In further including：
Third computing unit, for calculating the first denoising image and the Y-PSNR for waiting for denoising, by described first
Denoising image waits for denoising image as first, and triggers described image acquiring unit；When the variation of the Y-PSNR tends to
When stablizing, the optimal sparse coefficient of denoising image corresponding with the Y-PSNR is got, and trigger the second image reconstruction
Unit；
Second image reconstruction unit, the excessively complete dictionary for calculating the preset dictionary, according to the excessively complete dictionary
Pass through reconstruction of function x=D α with the optimal sparse coefficientyReconstruct obtains denoising image, wherein αyFor optimal sparse coefficient, D is
Cross complete dictionary.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810297729.8A CN108537752B (en) | 2018-03-30 | 2018-03-30 | Image processing method and device based on non-local self-similarity and sparse representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810297729.8A CN108537752B (en) | 2018-03-30 | 2018-03-30 | Image processing method and device based on non-local self-similarity and sparse representation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108537752A true CN108537752A (en) | 2018-09-14 |
| CN108537752B CN108537752B (en) | 2022-06-24 |
Family
ID=63483154
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810297729.8A Active CN108537752B (en) | 2018-03-30 | 2018-03-30 | Image processing method and device based on non-local self-similarity and sparse representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108537752B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109544477A (en) * | 2018-11-23 | 2019-03-29 | 南通大学 | Image denoising algorithm based on self-adapting dictionary study rarefaction representation |
| CN110971786A (en) * | 2019-11-21 | 2020-04-07 | 维沃移动通信有限公司 | Shooting method and electronic equipment |
| CN113642593A (en) * | 2020-04-27 | 2021-11-12 | 台达电子工业股份有限公司 | Image processing method and image processing system |
| CN115327289A (en) * | 2022-06-27 | 2022-11-11 | 华南理工大学 | Sparse signal power distribution network fault positioning method based on over-complete reconstruction representation |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102622735A (en) * | 2012-04-18 | 2012-08-01 | 江苏技术师范学院 | Secondary de-noising image processing method |
| US20130156340A1 (en) * | 2011-12-20 | 2013-06-20 | Fatih Porikli | Image Filtering by Sparse Reconstruction on Affinity Net |
| CN103971346A (en) * | 2014-05-28 | 2014-08-06 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image spot-inhibiting method based on spare domain noise distribution constraint |
| CN104091343A (en) * | 2014-07-22 | 2014-10-08 | 西北工业大学 | Image quality evaluation method based on sparse structure |
| CN104346787A (en) * | 2014-11-25 | 2015-02-11 | 成都卫士通信息产业股份有限公司 | Non-local mean image denoising algorithm |
| CN105279740A (en) * | 2015-11-17 | 2016-01-27 | 华东交通大学 | Image denoising method based on sparse regularization |
| US20160086317A1 (en) * | 2014-09-23 | 2016-03-24 | Intel Corporation | Non-local means image denoising with detail preservation using self-similarity driven blending |
| CN105513026A (en) * | 2015-12-17 | 2016-04-20 | 浙江大学 | Compressed sensing reconstruction method based on image nonlocal similarity |
| CN106157254A (en) * | 2015-04-21 | 2016-11-23 | 南京理工大学 | Rarefaction representation remote sensing images denoising method based on non local self-similarity |
| CN106169180A (en) * | 2016-07-13 | 2016-11-30 | 桂林电子科技大学 | A kind of non-local sparse based on group represents additive noise minimizing technology |
| CN106204483A (en) * | 2016-07-08 | 2016-12-07 | 桂林电子科技大学 | Multiplicative noise minimizing technology based on non local self-adapting dictionary |
| CN107067367A (en) * | 2016-09-08 | 2017-08-18 | 南京工程学院 | A kind of Image Super-resolution Reconstruction processing method |
-
2018
- 2018-03-30 CN CN201810297729.8A patent/CN108537752B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130156340A1 (en) * | 2011-12-20 | 2013-06-20 | Fatih Porikli | Image Filtering by Sparse Reconstruction on Affinity Net |
| CN102622735A (en) * | 2012-04-18 | 2012-08-01 | 江苏技术师范学院 | Secondary de-noising image processing method |
| CN103971346A (en) * | 2014-05-28 | 2014-08-06 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image spot-inhibiting method based on spare domain noise distribution constraint |
| CN104091343A (en) * | 2014-07-22 | 2014-10-08 | 西北工业大学 | Image quality evaluation method based on sparse structure |
| US20160086317A1 (en) * | 2014-09-23 | 2016-03-24 | Intel Corporation | Non-local means image denoising with detail preservation using self-similarity driven blending |
| CN104346787A (en) * | 2014-11-25 | 2015-02-11 | 成都卫士通信息产业股份有限公司 | Non-local mean image denoising algorithm |
| CN106157254A (en) * | 2015-04-21 | 2016-11-23 | 南京理工大学 | Rarefaction representation remote sensing images denoising method based on non local self-similarity |
| CN105279740A (en) * | 2015-11-17 | 2016-01-27 | 华东交通大学 | Image denoising method based on sparse regularization |
| CN105513026A (en) * | 2015-12-17 | 2016-04-20 | 浙江大学 | Compressed sensing reconstruction method based on image nonlocal similarity |
| CN106204483A (en) * | 2016-07-08 | 2016-12-07 | 桂林电子科技大学 | Multiplicative noise minimizing technology based on non local self-adapting dictionary |
| CN106169180A (en) * | 2016-07-13 | 2016-11-30 | 桂林电子科技大学 | A kind of non-local sparse based on group represents additive noise minimizing technology |
| CN107067367A (en) * | 2016-09-08 | 2017-08-18 | 南京工程学院 | A kind of Image Super-resolution Reconstruction processing method |
Non-Patent Citations (3)
| Title |
|---|
| CAI SHUTING ET AL: "Optimizing the framework of image denoising based on sparse and redundant representations", 《2016 CHINESE CONTROL AND DECISION CONFERENCE》 * |
| CAI SHUTING: "A Dictionary-Learning Algorithm Based on Method of Optimal Directions and Approximate K-SVD", 《PROCEEDINGS OF THE 35TH CHINESE CONTROL CONFERENCE》 * |
| 翁少佳: "基于稀疏表示的图像去噪算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109544477A (en) * | 2018-11-23 | 2019-03-29 | 南通大学 | Image denoising algorithm based on self-adapting dictionary study rarefaction representation |
| CN110971786A (en) * | 2019-11-21 | 2020-04-07 | 维沃移动通信有限公司 | Shooting method and electronic equipment |
| CN110971786B (en) * | 2019-11-21 | 2022-04-19 | 维沃移动通信有限公司 | Shooting method and electronic equipment |
| CN113642593A (en) * | 2020-04-27 | 2021-11-12 | 台达电子工业股份有限公司 | Image processing method and image processing system |
| CN113642593B (en) * | 2020-04-27 | 2024-03-01 | 台达电子工业股份有限公司 | Image processing method and image processing system |
| CN115327289A (en) * | 2022-06-27 | 2022-11-11 | 华南理工大学 | Sparse signal power distribution network fault positioning method based on over-complete reconstruction representation |
| CN115327289B (en) * | 2022-06-27 | 2024-04-16 | 华南理工大学 | Sparse signal power distribution network fault positioning method based on reconstruction overcomplete representation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108537752B (en) | 2022-06-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guo et al. | An efficient SVD-based method for image denoising | |
| Pu et al. | A fractional-order variational framework for retinex: fractional-order partial differential equation-based formulation for multi-scale nonlocal contrast enhancement with texture preserving | |
| CN107103326B (en) | Collaborative significance detection method based on super-pixel clustering | |
| CN108537752A (en) | Image processing method based on non local self-similarity and rarefaction representation and device | |
| Manap et al. | Non-distortion-specific no-reference image quality assessment: A survey | |
| Duval-Poo et al. | Edges and corners with shearlets | |
| CN103093434B (en) | Non-local wiener filtering image denoising method based on singular value decomposition | |
| CN103049892A (en) | Non-local image denoising method based on similar block matrix rank minimization | |
| CN110765882B (en) | Video tag determination method, device, server and storage medium | |
| CN103077506A (en) | Local and non-local combined self-adaption image denoising method | |
| Liu et al. | Adaptive sparse coding on PCA dictionary for image denoising | |
| CN110232318A (en) | Acupuncture point recognition methods, device, electronic equipment and storage medium | |
| Li et al. | Image super-resolution with parametric sparse model learning | |
| CN112712546A (en) | Target tracking method based on twin neural network | |
| Patel et al. | Separated component-based restoration of speckled SAR images | |
| CN112232134A (en) | Human body posture estimation method based on hourglass network and attention mechanism | |
| CN109410246B (en) | Visual tracking method and device based on correlation filtering | |
| CN110570394A (en) | medical image segmentation method, device, equipment and storage medium | |
| CN112907569A (en) | Head image area segmentation method and device, electronic equipment and storage medium | |
| CN110222734B (en) | Bayesian network learning method, intelligent device and storage device | |
| Bors et al. | Object classification in 3-D images using alpha-trimmed mean radial basis function network | |
| Li et al. | Irregular mask image inpainting based on progressive generative adversarial networks | |
| Li et al. | Bayesian multiscale smoothing of Gaussian noised images | |
| Liu et al. | Automated binocular vision measurement of food dimensions and volume for dietary evaluation | |
| Chen et al. | Noisy image segmentation based on wavelet transform and active contour model |
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 |