CN110349101A - A kind of multiple dimensioned united Wavelet image denoising method - Google Patents
A kind of multiple dimensioned united Wavelet image denoising method Download PDFInfo
- Publication number
- CN110349101A CN110349101A CN201910560473.XA CN201910560473A CN110349101A CN 110349101 A CN110349101 A CN 110349101A CN 201910560473 A CN201910560473 A CN 201910560473A CN 110349101 A CN110349101 A CN 110349101A
- Authority
- CN
- China
- Prior art keywords
- high frequency
- matrix
- coefficient
- low frequency
- sparse
- 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
- 238000000034 method Methods 0.000 title claims abstract description 82
- 239000011159 matrix material Substances 0.000 claims abstract description 123
- 230000009466 transformation Effects 0.000 claims abstract description 23
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 14
- 239000004615 ingredient Substances 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 6
- 238000003064 k means clustering Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 2
- 230000015572 biosynthetic process Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 10
- 238000012545 processing Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 8
- 238000006243 chemical reaction Methods 0.000 description 3
- 239000006002 Pepper Substances 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 235000002566 Capsicum Nutrition 0.000 description 1
- 235000016761 Piper aduncum Nutrition 0.000 description 1
- 235000017804 Piper guineense Nutrition 0.000 description 1
- 244000203593 Piper nigrum Species 0.000 description 1
- 235000008184 Piper nigrum Nutrition 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20064—Wavelet transform [DWT]
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The multiple dimensioned united Wavelet image denoising method of one kind disclosed by the invention, includes the following steps: S01: carrying out wavelet decomposition to image, obtains low frequency coefficient and high frequency coefficient under different scale;S02: high frequency coefficient and low frequency coefficient are split and joint of classifying, and constitute high frequency matrix and low frequency matrices;S03: high frequency is obtained by cluster and clusters matrix and low frequency cluster matrix;S04: pass through sparse acquisition high frequency sparse coefficient and low frequency sparse coefficient;S05: rebuilding high frequency coefficient and low frequency coefficient based on sparse coefficient to obtain denoising high frequency coefficient and denoising low frequency coefficient, carries out wavelet inverse transformation reconstruction image using denoising high frequency coefficient and denoising low frequency coefficient, to obtain denoising image.The multiple dimensioned united Wavelet image denoising method of one kind provided by the invention, the characteristics of making full use of wavelet transformation, enhance the separating effect of image effective information and noise, to realize higher denoising performance in combination with sparse method is clustered.
Description
Technical field
The present invention relates to field of image processings, particularly belong to a kind of multiple dimensioned united Wavelet image denoising method.
Background technique
Digital Image Processing, which refers to the process of, to be converted into digital signal for picture signal and is handled with computer it,
Purpose is processed to image information to meet the visual psychology of people or specific application demand.Image processing techniques is many
Field receives significant attention and achieves many pioneering achievements, such as aerospace, biomedical engineering, industrial detection, machine
People's vision, police and judicial, military guidance, culture and arts etc.;Since the mid-1970s, with computer technology and artificial intelligence
Can, the rapid development of cognitive science research, Digital Image Processing Xiang Genggao, deeper time develop.
The source numerous and complicated of picture noise, picture signal are easy in acquisition, compression, processing, transmission and reconstruction process
By various noise jammings, quality of image signals is caused to be degenerated, affect visual effect, for example common noise includes due to light
Flicker noise caused by photon fluctuation (shot noise) in electric signal conversion process, since the fuel factor of device circuitry generates
Thermal noise (thermal noise), further include the quantizing noise (quantization due to being generated in analog-digital conversion process
) and salt-pepper noise caused by component failure (Salt&Pepper noise) etc. noise.It is many important thin in image
Section information is covered by noise, seriously affects the subsequent applications and Development volue of image, so, image denoising is image processing techniques
In a Xiang Jiben and crucial technology, while being also one of the problem of field of image processing.
The difficult point of image denoising is that noise destroys raw image data, and denoising can only be based on the data being destroyed
Estimated, and combine some prior informations of image, obtain the estimation of clean image, on the other hand, due to being not aware that original
Beginning image data, that is, ideal noise-free picture data, the evaluation of the quality of image lack the method for unified standard, so, image
Even if noise-removed technology has studied decades by scientist, it is still the research paid close attention to and an opened heat now
Point.Existing Denoising Algorithm, it is simple easily to realize if airspace denoises (the methods of mean value, intermediate value), but image while reduction noise
Also more serious blurring effect is generated;Image is carried out Fourier transformation, is transformed into frequency domain from airspace by frequency domain filtering method,
Image is handled in frequency domain, then obtains airspace denoising image using inverse transformation, since the details and noise of image are in frequency
Domain shows as high-frequency characteristic, and when the two generates overlapping, denoising effect is undesirable.
Ineffective the having its source in of traditional Denoising Algorithm can not reasonably separate image information with noise.Small echo
Transformation chooses flexible etc. feature with its low entropy, multiresolution, decorrelation and transformation base, extensive in terms of image denoising
Using Image denoising method using wavelet main method can be divided into three classes: (1) modulus maximum, (2) Relativity of Coefficients method, (3) coefficient threshold
Method;Image denoising method using wavelet algorithm in effect compared to traditional algorithm for, be obviously improved really, still, wavelet algorithm
There are still some problems: modulus maximum and Relativity of Coefficients method are due to needing the wavelet coefficient under comprehensive different scale to complete
Tagsort realizes denoising, and algorithm complexity height denoises effect generally without being widely adopted;Coefficient threshold method is simple because of it
Efficient application is the most extensive, but the accuracy of threshold value selection is limited to the priori knowledge of wavelet coefficient, another aspect threshold method
Limited for the separating capacity of noise coefficient and signal coefficient, especially global threshold denoising method causes while denoising
The loss of image detail information generates image blurring effect.
Summary of the invention
The object of the present invention is to provide a kind of multiple dimensioned united Wavelet image denoising methods, make full use of wavelet transformation
Feature enhances the separating effect of image effective information and noise, to realize higher go in combination with sparse method is clustered
It makes an uproar performance.
To achieve the goals above, the present invention adopts the following technical scheme: a kind of multiple dimensioned united Image denoising method using wavelet
Method includes the following steps:
S01: it determines the decomposition scale of image, and wavelet decomposition is carried out to image, obtain the low frequency coefficient under different scale
And high frequency coefficient;
High frequency coefficient and low frequency coefficient: being split by S02 according to the tile size of setting, obtain high frequency coefficient block and
Low frequency coefficient block, and by under different scale high frequency coefficient block and low frequency coefficient block carry out classification joint, constitute high frequency matrix and
Low frequency matrices;
S03: and high frequency matrix and low frequency matrices are clustered, respectively obtain high frequency cluster matrix and low frequency cluster square
Battle array, wherein it includes cluster centre and corresponding ingredient that high frequency, which clusters matrix and low frequency cluster matrix,;
S04: matrix is clustered to high frequency respectively and low frequency cluster matrix progress is sparse, high frequency sparse coefficient is obtained and low frequency is dilute
Sparse coefficient;
S05: being based on high frequency sparse coefficient and low frequency sparse coefficient, rebuild respectively to high frequency coefficient and low frequency coefficient,
It obtains denoising high frequency coefficient and denoises low frequency coefficient, carry out wavelet inverse transformation weight using denoising high frequency coefficient and denoising low frequency coefficient
Image is built, to obtain denoising image.
Further, the high frequency coefficient block under different scale is carried out classifying to combine specifically including by the step S02: will be divided
High frequency coefficient block after cutting is divided into the horizontal coefficient block of high frequency, the vertical coefficient block of high frequency and high frequency diagonal linear system several piece three classes, and will
The horizontal coefficient block of the high frequency of different scale, the vertical coefficient block of high frequency and high frequency diagonal linear system several piece gather respectively, are formed high
The vertical matrix of frequency level matrix, high frequency and high frequency diagonal wire matrix.
Further, the low frequency coefficient block under the different scale after segmentation is carried out classification joint specifically by the step S02
It include: that the low frequency coefficient set of blocks of different scale is formed into low frequency matrices.
Further, the high frequency cluster matrix is based on image and forms high frequency cluster index matrix, and low frequency cluster index
Matrix is identical as the high frequency cluster index matrix.
Further, the step S03 is specifically included:
S031: clustering high frequency matrix, obtains high frequency cluster matrix;
S032: the high frequency cluster matrix is based on image and forms high frequency cluster index matrix;
S033: low frequency cluster index matrix is made to be equal to high frequency cluster index matrix;
S034: the low frequency cluster index matrix is based on image and show that low frequency clusters matrix.
Further, method high frequency matrix clustered in the step S031 include one of following methods or
It is a variety of: mean shift clustering algorithm, DBSCAN clustering algorithm, K-means clustering algorithm.
Further, the step S04 is carried out sparse using principal component analytical method or dictionary learning method.
Further, the step S04 is carried out sparse using dictionary learning method, is specifically included:
S041: it is obtained in high frequency cluster matrix and low frequency cluster matrix respectively in each cluster using the method for dictionary learning
The sparse dictionary of the heart;
S042: corresponding sparse dictionary is respectively adopted to the corresponding each ingredient of cluster centre each in high frequency cluster matrix
It carries out sparse, obtains high frequency sparse coefficient;Corresponding sparse dictionary is respectively adopted to each cluster centre in low frequency cluster matrix
Corresponding each ingredient progress is sparse, obtains low frequency sparse coefficient.
Further, the step S05 is using compressed sensing based sparse reconstruction method to high frequency coefficient and low frequency system
Number is rebuild.
Further, the compressed sensing based sparse reconstruction method is that convex optimization is built.
The invention has the benefit that one aspect of the present invention combines similar spy between the cluster of wavelet transformation and scale
Point promotes the whole Clustering Effect of picture structure;On the other hand, the characteristics of invariance sparse based on noise, using sparse reconstruction
Method separate picture and noise, finally sparse coefficient is rebuild using compressed sensing based sparse reconstruction method, and
Denoising image is obtained by wavelet inverse transformation.The method that the present invention combines sparse reconstruction using wavelet transformation, enhancing image are effective
The separation degree of information and noise, and then achieve the purpose that promote image denoising effect.
Detailed description of the invention
Attached drawing 1 is a kind of schematic diagram of multiple dimensioned united Wavelet image denoising method of the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, with reference to the accompanying drawing to specific reality of the invention
The mode of applying is described in further detail.
As shown in Fig. 1, the multiple dimensioned united Wavelet image denoising method of one kind provided by the invention, including walk as follows
It is rapid:
S01: it determines the decomposition scale of image, and wavelet decomposition is carried out to image, obtain the low frequency coefficient under different scale
And high frequency coefficient.Different scale, that is, wavelet transformation different decomposition scale, wavelet transformation has the resolution characteristic of different scale, small
The degree that wave conversion decomposes image under different scale is different.
Wherein, wavelet decomposition is carried out to image, it can be according to the suitable wavelet basis of image selection, such as existing basic small echo
Base Haar small echo, Db series, the Sym series of near symmetrical and Coif series, dual density wavelet basis, double tree wavelet basis and shape are small
The selection of wave base etc., wavelet basis can be selected according to real image feature.Figure is carried out by taking Haar wavelet basis as an example below
As illustrating for wavelet decomposition, the small echo based structures are simple, and wavelet transformation complexity is low, using extensive.Specific decomposable process
It is as follows: to determine the decomposition scale j of image, then the transformation coefficient of available wavelet field, wherein wavelet field is a kind of name in a name space
Claim, wavelet field can be image by space existing after wavelet transformation, obtain frequency domain later similar to Fourier transformation;
Or wavelet field can refer to the space where the wavelet coefficient that each Scale Decomposition obtains;
The low frequency coefficient of each scale:
The high frequency coefficient of each scale:
Wherein M, N are the size of original image, and m, n are the component index after wavelet transformation,WithFor coefficient of the image at wavelet decomposition scales j, whereinFor low frequency coefficient,For the high frequency coefficient in the direction horizontal H, vertical V and diagonal line D,It is respectively low with ψ (x, y)
Logical and high pass resolution filter.Original input picture f (x, y) is the low-pass component under the highest resolution expression of image, i.e.,
High frequency coefficient and low frequency coefficient: being split by S02 according to the tile size of setting, obtain high frequency coefficient block and
Low frequency coefficient block, and by under different scale high frequency coefficient block and low frequency coefficient block carry out classification joint, constitute high frequency matrix and
Low frequency matrices.
Wherein, the image block of setting refers to the image block radius being set in advance, if setting image block radius as s, that is, schemes
As centered on each pixel, radius is that the square of s is split, i.e., the image block of (2s+1) * (2s+1), s are positive integer.?
When handling image data, usually we whole to single image progress will not be handled simultaneously, but by picture breakdown at fixed big
Small fritter, i.e. image block, and each fritter is respectively processed, the processing result for then integrating each image block again obtains
The processing result of entire image.The above-mentioned high frequency coefficient obtained under each scale and low frequency coefficient are one or more larger
Matrix data, need to be splitted into fritter, i.e. high frequency coefficient block or low frequency coefficient block, be respectively processed.
Specifically, carrying out classification to high frequency coefficient block to combine including: that the high frequency coefficient block after segmentation is divided into high frequency level
The vertical coefficient block of coefficient block, high frequency and high frequency diagonal linear system several piece three classes, and by the horizontal coefficient block of the high frequency of different scale, high frequency
Vertical coefficient block and high frequency diagonal linear system several piece gather respectively, be respectively formed the vertical matrix of high frequency level matrix, high frequency and
High frequency diagonal wire matrix, triple combination get up to be collectively formed high frequency matrix
To low frequency coefficient blockHorizontal, vertical and diagonal is needed not distinguish between, is directly joined together:
The low frequency coefficient set of blocks of different scale after segmentation is got up, low frequency matrices are formed
The high frequency matrix (the vertical matrix of high frequency level matrix, high frequency and high frequency diagonal wire matrix) in above three direction is same
Size is identical under one scale, high frequency matrix is formed after fusion, and identical as the low frequency matrices size under the scale.
S03: clustering high frequency matrix and low frequency matrices, respectively obtains high frequency cluster matrix and low frequency cluster matrix,
Wherein, high frequency cluster matrix includes cluster centre and corresponding ingredient, low frequency cluster matrix include cluster centre and it is corresponding at
Point.Here, cluster centre is preset, wherein the quantity of cluster centre is determining, the specific side for determining cluster centre quantity
Method can be carried out using determining method in the prior art;But the position of cluster centre is not optimal, and it is common
Clustering algorithm can all be finally reached the purpose of cluster centre optimization by iteration, so the cluster centre number after cluster
Constant, position is optimal, and cluster matrix is collectively formed in all cluster centres ingredient corresponding with around it.Here gather
The corresponding ingredient in class center is the above-mentioned high frequency coefficient block or low frequency coefficient block referred to, and each image block or coefficient block are corresponding
For a kind of ingredient, in cluster process, similar one kind image block or coefficient block are aggregated together, and form a cluster centre
And corresponding ingredient.
Wherein, high frequency cluster matrix, which is based on image, can form high frequency cluster index matrix, i.e. high frequency clusters matrix in image
In index;Low frequency cluster matrix is based on image can form low frequency cluster index matrix, i.e., low frequency cluster matrix is in the picture
Index;Matrix is clustered in the present invention and cluster index matrix is two kinds of expression ways of the same content, is mutually to convert
, i.e., if having obtained cluster index matrix, cluster matrix can be shown that vice versa by index matrix.Above-mentioned high frequency
The pixel that matrix and low frequency matrices are included corresponds, and the high frequency coefficient block and low frequency coefficient block one after segmentation are a pair of
It answers, high frequency clusters matrix and low frequency cluster matrix corresponds, and high frequency cluster index matrix and low frequency cluster index matrix are one by one
It is corresponding.
Square is clustered since the effective information (edge, details etc.) of image embodies in high frequency matrix, therefore according to high frequency
Battle array in cluster centre and corresponding ingredient it is more representative to the classification of image, under same scale, single high frequency matrix with
Low frequency matrices size is identical, so low frequency cluster matrix directlys adopt high frequency cluster index matrix as cluster result.Specifically adopt
It is clustered with such as under type:
S031: clustering high frequency matrix, obtains high frequency cluster matrix;Wherein, high frequency cluster matrix includes in cluster
The heart and corresponding ingredient, cluster centre here have fixed quantity (presetting) and optimal position (cluster iteration
It obtains).Specific clustering method can be selected according to characteristics of image and distribution in practical applications, such as can be mean value
Deviate clustering algorithm, DBSCAN clustering algorithm, K-means clustering algorithm etc..The determination method of specific cluster centre number
It can be carried out using determining method in the prior art.
S032: high frequency clusters matrix and is based on image formation high frequency cluster index matrix.Specifically high frequency cluster index matrix isWherein Cls is high frequency cluster index matrix, and center is the cluster of high frequency matrix
Center, pat_num are the index of the corresponding ingredient of the cluster centre in the picture, H, V, and D respectively indicates horizontal direction, vertical side
To and diagonal.That is, the cluster centre of high frequency matrix is remained unchanged, at the same by the cluster centre it is corresponding at
Point be converted to its index in the picture you can get it high frequency cluster index matrix.High frequency cluster index is only gived in this step
The expression formula of matrix, and high frequency cluster matrix does not provide specific expression-form;It is to be understood that high frequency cluster matrix is high
Result after frequency matrix is clustered describes;High frequency cluster index matrix is the specific manifestation after high frequency matrix is clustered
Form.
S033: low frequency cluster index matrix is made to be equal to high frequency cluster index matrix.
S034: low frequency cluster index matrix is based on image and show that low frequency clusters matrix, and low frequency cluster matrix at this time has
Cluster centre identical with high frequency cluster matrix, and the corresponding ingredient of each cluster centre corresponds.
S04: matrix is clustered to high frequency respectively and low frequency cluster matrix progress is sparse, high frequency sparse coefficient is obtained and low frequency is dilute
Sparse coefficient.During carrying out rarefaction representation to image, noisy image can be divided into useful information and noise information two parts, figure
Useful information has certain structure as in, and noise is random and without specific structure, and useful information can obtain most sparse
It indicates, and noise does not have this characteristic, so, it can be by noise remove, to realize during solving rarefaction representation
Denoising.
Specific Sparse methods in practical applications can be according to the selection for carrying out Sparse methods the characteristics of real image, example
It is sparse such as to can choose the methods of principal component analysis or dictionary learning progress;Below by the rarefaction method pair of dictionary learning
The Thinning Process is illustrated, and is specifically included:
S041: it is based on cluster resultIt is obtained respectively using the method for dictionary learning
Obtain the sparse dictionary of each cluster centre in high frequency cluster matrix and low frequency cluster matrix, that is, sparse basis;I.e. each is poly-
Class center can be obtained a sparse dictionary;
S042: corresponding sparse dictionary is respectively adopted to the corresponding each ingredient of cluster centre each in high frequency cluster matrix
It carries out sparse, obtains high frequency sparse coefficient;Corresponding sparse dictionary is respectively adopted to each cluster centre in low frequency cluster matrix
Corresponding each ingredient progress is sparse, obtains low frequency sparse coefficient.Since Thinning Process is corresponding for each cluster centre
What each ingredient carried out, therefore obtain to be the corresponding image block of sparse coefficient or coefficient block.
S05: being based on high frequency sparse coefficient and low frequency sparse coefficient, rebuild respectively to high frequency coefficient and low frequency coefficient,
It obtains denoising high frequency coefficient and denoises low frequency coefficient, carry out wavelet inverse transformation weight using denoising high frequency coefficient and denoising low frequency coefficient
Image is built, to obtain denoising image.
Wherein, specific sparse coefficient is rebuild can select suitable method for reconstructing according to real image feature, including but
It is not limited to compressed sensing based sparse reconstruction method, specific compressed sensing based sparse reconstruction method can use convex excellent
Change method is weighed into convexity optimization is specifically as follows L1 norm optimization, LpNorm optimization (0 < p < 1), nuclear norm optimization etc..
By taking L1 norm optimization's method as an example: firstly, the high frequency sparse coefficient and low frequency sparse coefficient of the image block based on each scale, using L1
The high frequency coefficient and low frequency coefficient of each scale of method reconstruction image of norm optimization obtain denoising high frequency coefficient and denoise low frequency system
Then then number carries out wavelet inverse transformation completion according to the denoising high frequency coefficient of each scale of reconstruction and denoising low frequency coefficient and goes
Image reconstruction of making an uproar work, the image after being denoised.
The present invention is based on the high frequency coefficients under each scale of wavelet decomposition transform to have the characteristics that cluster is similar with low frequency coefficient,
The high frequency coefficient and low frequency coefficient for combining each scale by combining there is the wavelet coefficient of similitude image information is added
By force, compressed sensing based rarefaction reconstruction side is finally then used using the further separate picture of sparse method and noise is clustered
Method rebuilds sparse coefficient, and obtains denoising image by wavelet inverse transformation.Inventive method makes full use of wavelet transformation
Feature enhances the separating effect of image effective information and noise, to realize higher go in combination with sparse method is clustered
It makes an uproar performance.
The above description is only a preferred embodiment of the present invention, and the embodiment is not intended to limit protection model of the invention
It encloses, therefore all with the variation of equivalent structure made by specification and accompanying drawing content of the invention, similarly should be included in this hair
In the protection scope of bright appended claims.
Claims (10)
1. a kind of multiple dimensioned united Wavelet image denoising method, which comprises the steps of:
S01: it determines the decomposition scale of image, and wavelet decomposition is carried out to image, obtain low frequency coefficient and height under different scale
Frequency coefficient;
High frequency coefficient and low frequency coefficient: being split by S02 according to the tile size of setting, obtains high frequency coefficient block and low frequency
Coefficient block, and by under different scale high frequency coefficient block and low frequency coefficient block carry out classification joint, constitute high frequency matrix and low frequency
Matrix;
S03: clustering high frequency matrix and low frequency matrices, respectively obtains high frequency cluster matrix and low frequency cluster matrix, wherein
It includes cluster centre and corresponding ingredient that high frequency, which clusters matrix and low frequency cluster matrix,;
S04: matrix is clustered to high frequency respectively and low frequency cluster matrix carries out sparse, acquisition high frequency sparse coefficient and the sparse system of low frequency
Number;
S05: it is based on high frequency sparse coefficient and low frequency sparse coefficient, high frequency coefficient and low frequency coefficient are rebuild respectively, obtained
High frequency coefficient and denoising low frequency coefficient are denoised, wavelet inverse transformation is carried out using denoising high frequency coefficient and denoising low frequency coefficient and rebuilds figure
Picture, to obtain denoising image.
2. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 1, which is characterized in that the step
High frequency coefficient block under different scale is carried out classifying to combine specifically including by S02: the high frequency coefficient block after segmentation is divided into high frequency
The vertical coefficient block of horizontal coefficients block, high frequency and high frequency diagonal linear system several piece three classes, and by the horizontal coefficient block of the high frequency of different scale,
The vertical coefficient block of high frequency and high frequency diagonal linear system several piece gather respectively, formed high frequency level matrix, the vertical matrix of high frequency and
High frequency diagonal wire matrix.
3. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 1, which is characterized in that the step
Low frequency coefficient block under different scale is carried out classifying to combine specifically including by S02: by the low frequency coefficient of the different scale after segmentation
Set of blocks gets up, and forms low frequency matrices.
4. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 1, which is characterized in that the high frequency
It clusters matrix and is based on image formation high frequency cluster index matrix, and low frequency cluster index matrix and the high frequency cluster index matrix
It is identical.
5. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 4, which is characterized in that the step
S03 is specifically included:
S031: clustering high frequency matrix, obtains high frequency cluster matrix;
S032: the high frequency cluster matrix is based on image and forms high frequency cluster index matrix;
S033: low frequency cluster index matrix is made to be equal to high frequency cluster index matrix;
S034: the low frequency cluster index matrix is based on image and show that low frequency clusters matrix.
6. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 5, which is characterized in that the step
The method clustered in S031 to high frequency matrix includes one of following methods or a variety of: mean shift clustering algorithm,
DBSCAN clustering algorithm, K-means clustering algorithm.
7. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 1, which is characterized in that the step
S04 is carried out sparse using principal component analytical method or dictionary learning method.
8. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 7, which is characterized in that the step
S04 is sparse using dictionary learning method progress, specifically includes:
S041: each cluster centre in high frequency cluster matrix and low frequency cluster matrix is obtained using the method for dictionary learning respectively
Sparse dictionary;
S042: corresponding sparse dictionary is respectively adopted, the corresponding each ingredient of cluster centre each in high frequency cluster matrix is carried out
It is sparse, obtain high frequency sparse coefficient;It is corresponding to each cluster centre in low frequency cluster matrix that corresponding sparse dictionary is respectively adopted
Each ingredient carry out it is sparse, obtain low frequency sparse coefficient.
9. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 1, which is characterized in that the step
S05 rebuilds high frequency coefficient and low frequency coefficient using compressed sensing based sparse reconstruction method.
10. the multiple dimensioned united Wavelet image denoising method of one kind according to claim 9, which is characterized in that the base
In compressed sensing sparse reconstruction method be convex optimization.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910560473.XA CN110349101B (en) | 2019-06-26 | 2019-06-26 | Multi-scale combined wavelet image denoising method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910560473.XA CN110349101B (en) | 2019-06-26 | 2019-06-26 | Multi-scale combined wavelet image denoising method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110349101A true CN110349101A (en) | 2019-10-18 |
CN110349101B CN110349101B (en) | 2023-11-14 |
Family
ID=68183138
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910560473.XA Active CN110349101B (en) | 2019-06-26 | 2019-06-26 | Multi-scale combined wavelet image denoising method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110349101B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112613583A (en) * | 2021-01-05 | 2021-04-06 | 广东工业大学 | High-frequency information extraction clustering method for low-frequency noise face image |
CN114758017A (en) * | 2022-04-24 | 2022-07-15 | 启东市恒通橡胶制品厂(普通合伙) | Compression transmission method for detecting abnormity of rubber sealing ring |
CN115389888A (en) * | 2022-10-28 | 2022-11-25 | 山东科华电力技术有限公司 | Partial discharge real-time monitoring system based on high-voltage cable |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105869126A (en) * | 2016-04-05 | 2016-08-17 | 天津大学 | Compression perception denoising method of high-spectral images |
US20160284067A1 (en) * | 2013-11-08 | 2016-09-29 | Huawei Device Co., Ltd. | Image Denoising Method and Terminal |
CN108491868A (en) * | 2018-03-12 | 2018-09-04 | 广东工业大学 | A kind of image processing method and device based on k-means cluster and dictionary learning |
-
2019
- 2019-06-26 CN CN201910560473.XA patent/CN110349101B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160284067A1 (en) * | 2013-11-08 | 2016-09-29 | Huawei Device Co., Ltd. | Image Denoising Method and Terminal |
CN105869126A (en) * | 2016-04-05 | 2016-08-17 | 天津大学 | Compression perception denoising method of high-spectral images |
CN108491868A (en) * | 2018-03-12 | 2018-09-04 | 广东工业大学 | A kind of image processing method and device based on k-means cluster and dictionary learning |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112613583A (en) * | 2021-01-05 | 2021-04-06 | 广东工业大学 | High-frequency information extraction clustering method for low-frequency noise face image |
CN112613583B (en) * | 2021-01-05 | 2023-07-21 | 广东工业大学 | High-frequency information extraction clustering method for low-frequency noise face image |
CN114758017A (en) * | 2022-04-24 | 2022-07-15 | 启东市恒通橡胶制品厂(普通合伙) | Compression transmission method for detecting abnormity of rubber sealing ring |
CN114758017B (en) * | 2022-04-24 | 2023-09-15 | 青岛仁盛新材料有限公司 | Compression transmission method for detecting abnormality of rubber sealing ring |
CN115389888A (en) * | 2022-10-28 | 2022-11-25 | 山东科华电力技术有限公司 | Partial discharge real-time monitoring system based on high-voltage cable |
CN115389888B (en) * | 2022-10-28 | 2023-01-31 | 山东科华电力技术有限公司 | Partial discharge real-time monitoring system based on high-voltage cable |
Also Published As
Publication number | Publication date |
---|---|
CN110349101B (en) | 2023-11-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Shih | Image processing and pattern recognition: fundamentals and techniques | |
CN110349101A (en) | A kind of multiple dimensioned united Wavelet image denoising method | |
CN106981057B (en) | RPCA-based NSST image fusion method | |
CN113362250B (en) | Image denoising method and system based on dual-tree quaternary wavelet and deep learning | |
JP2000508488A (en) | System and method for multi-resolution conversion of digital image information | |
US20020114464A1 (en) | Method for lighting- and view -angle-invariant face description with first- and second-order eigenfeatures | |
Muneeswaran et al. | Enhanced image compression using fractal and tree seed-bio inspired algorithm | |
Sharma et al. | From pyramids to state‐of‐the‐art: a study and comprehensive comparison of visible–infrared image fusion techniques | |
CN111915518A (en) | Hyperspectral image denoising method based on triple low-rank model | |
Cai et al. | Binarized spectral compressive imaging | |
Kaarna et al. | Compression of spectral images | |
WO2023241188A1 (en) | Data compression method for quantitative remote sensing application of unmanned aerial vehicle | |
CN116452450A (en) | Polarized image defogging method based on 3D convolution | |
CN114821239B (en) | Method for detecting plant diseases and insect pests in foggy environment | |
CN106780398B (en) | A kind of image de-noising method based on noise prediction | |
CN112991194B (en) | Infrared thermal wave image deblurring method based on depth residual error network | |
Mansoor et al. | On image compression using digital curvelet transform | |
Kaeli | Computational strategies for understanding underwater optical image datasets | |
Chen et al. | Texture Feature Extraction Method for Ground Nephogram Based on Contourlet and the Power Spectrum Analysis Algorithm. | |
Gao et al. | Multispectral image restoration via inter-and intra-block sparse estimation based on physically-induced joint spatiospectral structures | |
Talebi et al. | Asymptotic performance of global denoising | |
CN112950519B (en) | Novel infrared and visible light image fusion algorithm | |
CN113449737B (en) | Single-probe acoustic imaging method and device based on self-encoder | |
Hilles | Spatial Frequency Filtering Using Sofm For Image Compression | |
CN112200247B (en) | Image processing system and method based on multi-dimensional image mapping |
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 |