CN108734669A - Image denoising method based on wavelet transformation Wiener filtering and edge detection - Google Patents
Image denoising method based on wavelet transformation Wiener filtering and edge detection Download PDFInfo
- Publication number
- CN108734669A CN108734669A CN201710269122.4A CN201710269122A CN108734669A CN 108734669 A CN108734669 A CN 108734669A CN 201710269122 A CN201710269122 A CN 201710269122A CN 108734669 A CN108734669 A CN 108734669A
- Authority
- CN
- China
- Prior art keywords
- image
- wavelet
- formula
- wavelet decomposition
- wiener
- 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.)
- Pending
Links
- 230000009466 transformation Effects 0.000 title claims abstract description 33
- 238000001914 filtration Methods 0.000 title claims abstract description 29
- 238000003708 edge detection Methods 0.000 title claims abstract description 15
- 238000000034 method Methods 0.000 title claims abstract description 15
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 22
- 230000003044 adaptive effect Effects 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims abstract description 6
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000011946 reduction process Methods 0.000 abstract 1
- 238000005286 illumination Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 5
- 230000003068 static effect Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000029305 taxis Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/10—Image enhancement or restoration by non-spatial domain filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration by the use of local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- 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]
Abstract
The present invention provides a kind of image denoising method based on wavelet transformation Wiener filtering and edge detection, includes the following steps：Step 1, two layers of wavelet decomposition is carried out to original twilight image, extracts horizontal component, vertical component, the diagonal components of the approximation coefficient of second layer wavelet decomposition and the detail coefficients of second layer wavelet decomposition；Step 2, suitable filter window is chosen, noise reduction process is carried out to picture by improved adaptive wiener filter noise reduction algorithm；Step 3, edge detection is carried out by canny operators on the basis of artwork, the edge detected is overlapped with the reconstructed image after Wiener filtering, obtains final noise-reduced image.
Description
Technical field
It is especially a kind of based on wavelet transformation Wiener filtering and edge detection the present invention relates to a kind of image processing techniques
Image denoising method.
Background technology
Under poor light condition, since illumination is relatively low and the limitation of detector sensitivity etc., the obtained video image letter of system
It makes an uproar relatively low, influences eye-observation, or even can not effectively obtain target scene image.The noise of low-light video is mainly by CCD institutes
The quantum noise caused by the white noise and image intensifier of Gaussian Profile that meets generated is formed.
Twilight image noise reduction algorithm is broadly divided into airspace filter and transform domain filters two classes, and image transform domain denoising method is
Certain transformation is carried out to image, image is changed into transform domain from transform of spatial domain, then handle the transformation coefficient in transform domain,
Inverse transformation is carried out again, and image is transformed into spatial domain to achieve the purpose that remove image throat sound from transform domain.By image from spatial domain
There are many transform method for being transformed into transform domain, as Fourier transform, Walsh-Hadamard transform, cosine transform, Karhunen-Loeve transformation with
And wavelet transformation etc..And Fourier transform and wavelet transformation are then the transform methods for being commonly used in image denoising.Airspace filter
It is directly to carry out data operation on original image, the gray value of pixel is handled.Common spatial domain Image denoising algorithm
There are neighborhood averaging, medium filtering, low-pass filtering etc..Spatial domain is used alone or transform domain filtering is difficult to reducing picture noise
While retain image edge information.Therefore, this paper presents a kind of on the basis of time-domain filtering carry out airspace filter when
The low-light video filtering method of spatial domain mixing.
Invention content
The low-light of the purpose of the present invention is to provide a kind of improved Wiener filtering and edge detection based on wavelet transformation
Image denoising method, the present invention include the following steps：
Step 1, two layers of wavelet decomposition is carried out to original twilight image, extract second layer wavelet decomposition approximation coefficient and
Horizontal component, vertical component, the diagonal components of the detail coefficients of second layer wavelet decomposition；
Step 2, suitable filter window is chosen, picture is dropped by improved adaptive wiener filter noise reduction algorithm
It makes an uproar processing；
Step 3, edge detection is carried out by canny operators on the basis of artwork, by the edge detected with by tieing up
Filtered reconstructed image of receiving is overlapped, and obtains final noise-reduced image.
The present invention changes transform of spatial domain into wavelet field, and it is adaptive to the progress of noisy acoustic image to choose the suitable wavelet transformation number of plies
Wiener filtering denoising is answered, while noise-reduced image edge is waited for using the detection of log edge detection operators, image is divided into marginal zone and is waited for
Edge finally is superimposed to obtain final noise-reduced image by noise reduction area with the image noise reduction area after wavelet reconstruction.
The invention will be further described with reference to the accompanying drawings of the specification.
Description of the drawings
Fig. 1 is that the present invention is based on the twilight image noise reduction algorithm streams of the adaptive wiener filter of wavelet transformation and edge detection
Cheng Tu.
Fig. 2 is the Wiener filtering algorithm schematic diagram based on wavelet transformation.
Fig. 3 is that people stands on low-light level imaging experiment schematic diagram in the woods under 0.10lux illumination.
Fig. 4 is that static building tests schematic diagram under 0.22lux illumination.
Fig. 5 is that low-light moves portrait experiment schematic diagram under 0.22lux illumination.
Fig. 6 is that the static building of low-light tests schematic diagram under 0.5lux illumination.
Specific implementation mode
Step 1, wavelet transformation is carried out to original twilight image；
Assuming that pending graphical representation is Y=X+N；Wherein, X indicates the image of " clean ", and N is orthogonal, equal with X
Value is zero, variance σ2Gaussian noise matrix.It can be obtained after carrying out wavelet transformation：Y=x+n, y=WY, x=WX, n=in formula
WN, W are the transformation matrix of wavelet transformation.By the orthogonality of wavelet transformation it is found that n is still mean value is zero, variance σ2Height
This variable, and it is orthogonal with x.
Step 2, to pending image with ' db2 ' wavelet function carries out two layers of wavelet decomposition, and extracts the 2nd layer of wavelet decomposition
Approximation coefficient (low-frequency component), the detail coefficients horizontal component of the 2nd layer of wavelet decomposition, the detail coefficients of the 2nd layer of wavelet decomposition
Vertical component, the detail coefficients diagonal components of the 2nd layer of wavelet decomposition.
The horizontal component, vertical component and diagonal components for waiting for noise-reduced image Jing Guo wavelet decomposition is carried out adaptively respectively
Wiener filtering.
Algorithm steps are as shown in Figure 2.
According to institute in step 1 it is assumed that pending graphical representation is Y=X+N；It can be obtained after carrying out wavelet transformation：Y=x+n, formula
Middle y=WY, x=WX, n=WN, W are the transformation matrix of wavelet transformation.By the orthogonality of wavelet transformation it is found that n is still mean value
It is zero, variance δ2Gaussian variable, and it is orthogonal with x.Due to the decorrelation of orthogonal wavelet transformation, signal X is by just
Hand over the later output x of wavelet transformation be also it is incoherent, therefore, it is also believed that x is the Gaussian random variable of zero-mean, this
When, the form of Wiener filtering can be reduced to
Wherein E [] indicates the mathematic expectaion of variable, and a is Wiener filtering coefficient,Estimate for the optimum linearity of x.It considers
N is orthogonal with x, so having
E[x2]=E [y2]-δ2 (2)
E [y can be acquired one by one2], for being located at the E [y at coordinate (i, j)2], using yi,jAnd surrounding value estimation
It obtains.In order to without loss of generality, utilize the square window of one (2R+1) × (2R+1) (center of the window is located at coordinate (i, j))
In to yi,jValue averagely acquires
Wherein Qi,jIt is the pixel inner product and q in square window by wavelet transformationi,jIt is inner product and mean value, M is square window
Size.By qi,jAsApproximate evaluation value, i.e.,Substitute into (2) Shi Ke get
WhereinIt indicatesMathematic expectaion, qi,jIt isThe approximation of mathematic expectaion, δ2It is noise variance.Generation again
Enter (1) formula, each coefficient that can obtain Wiener filtering is
By E [x2This constraints of] >=0, so (6) formula can be write as max (qi,j-δ2,0)。
The least mean-square error expectation for defining Wiener filtering now is as follows：
LMSE=E { [xi,j-ai,jyi,j]2} (7)
Enable bi,j=δ2/qi,j, and can be obtained by (6)
LSME=E { [xi,j-(1-bi,j)(xi,j-ni,j)]2} (8)
To put it more simply, ignoring each subscript, (8) formula can be rewritten as again
LSME=E { [b (x+n)-n]2} (9)
When square window size is sufficiently large, it is believed that b is independent from each other with x and n.This hypothesis can receive completely.
Above formula is unfolded, (9) is substituted into according to the property of mathematic expectaion, and by (5) formula, can obtain
LMSE=E [b2]q+δ2{1-2E[b]} (10)
Assuming that if working as：
LMSE > E [x2] (11)
Wherein LMSE is that the least mean-square error of Wiener filtering it is expected that E [] is the mathematic expectaion of variable.Then for x's
Linear Estimation can become very poor, so solution relatively good at this time is that x is directly assigned 0.
BySubstitute into (7) Shi Ke get
The inequality of solution above, can obtain
Q < k δ2 (13)
WhereinSince the M representative values taken are 9 (image blocks of 3 × 3 pixels), the 49 (images of 7 × 7 pixels
Block), so k is 1.47,1.20.I other words when q is less than k δ2When, the Linear Estimation to x is abandoned, it is thus right directly by x taxes 0
Image completes the Wiener filtering algorithm based on wavelet decomposition.
It step 2-2, will be after the low frequency component of the wavelet transformation of noise-reduced image and each side after adaptive wiener filter
It is overlapped to component, using ' db2 ' wavelet function reconstructs to obtain the image after noise reduction.
Step 3:Using Canny operators to source images carry out edge detection, by the edge image detected with noise reduction after
Image is overlapped, and obtains final noise-reduced image.
Step 3-1, Canny operators have many advantages, such as low error rate, high position precision and inhibit false edge.In program
In, gray scale twilight image is read in first, and edge detection is carried out using Canny operators.When carrying out edge detection, using function
Automatically the threshold value calculated, and the threshold value is returned, the threshold value thresh that function returns is the threshold value after normalization, is obtained to noisy original
Figure carries out the edge image of edge extracting.
Step 3-2, edge image is overlapped to obtain final denoising figure with the image after Wavelet Domain Wiener Filtering
Picture.
Embodiment
Multiple experiment scenes shooting twilight images are had chosen herein, and reality has been carried out to the image shot in the case of a variety of respectively
Verification.Fig. 3 be under 0.10lux illumination people stand in the woods low-light level imaging experimental image handle sectional drawing, Fig. 4 be
Static building low-light level imaging experimental image processing sectional drawing, Fig. 5 are that low-light moves under 0.22lux illumination under 0.22lux illumination
Portrait experimental image handles sectional drawing, Fig. 6 is that the static building experimental image of low-light handles sectional drawing under 0.5lux illumination.
It can be seen that a kind of low-light of adaptive wiener filter and edge detection based on wavelet transformation proposed by the present invention
Image noise reduction algorithm overcomes single time-space domain filtering can be by the important high-frequency information such as image border while filtering out noise
The shortcomings that filtering out extracts the edge of noisy acoustic image while being filtered, restrained effectively twilight image
Noise so that image object and background profile are apparent, the acquisition for information of being more convenient for.
Claims (3)
1. a kind of image denoising method based on wavelet transformation Wiener filtering and edge detection, which is characterized in that including following step
Suddenly：
Step 1, two layers of wavelet decomposition is carried out to original twilight image, extracts the approximation coefficient and second of second layer wavelet decomposition
Horizontal component, vertical component, the diagonal components of the detail coefficients of layer wavelet decomposition；
Step 2, suitable filter window is chosen, picture is carried out at noise reduction by improved adaptive wiener filter noise reduction algorithm
Reason；
Step 3, edge detection is carried out by canny operators on the basis of artwork, the edge detected is filtered with by wiener
Reconstructed image after wave is overlapped, and obtains final noise-reduced image.
2. according to the method described in claim 1, it is characterized in that, the detailed process of step 1 is：
Step 1.1, wavelet transformation is carried out for original twilight image Y=X+N and obtains y+x+n, wherein X is clean image, N be with
X is uncorrelated and mean value be 0 variance is δ2Gaussian noise matrix, y=WY, x=WX, n=WN, W be wavelet transformation transformation square
Battle array, n are that uncorrelated and mean value is 0 to x and variance is δ2Gaussian variable；
Step 1.2, two layers of wavelet decomposition is carried out using db2 wavelet functions, and extracts the approximation coefficient of the 2nd layer of wavelet decomposition, the 2nd
The detail coefficients horizontal component of layer wavelet decomposition, the detail coefficients vertical component of the 2nd layer of wavelet decomposition, the 2nd layer of wavelet decomposition
Detail coefficients diagonal components.
3. according to the method described in claim 2, it is characterized in that, the detailed process of step 2 is：
Step 3.1, Wiener filtering function is set
Wherein E [] indicates the mathematic expectaion of variable, and a is Wiener filtering coefficient,Estimate for the optimum linearity of x, setting constraint item
Part E [x2]≥0；
Step 3.2, function is established
E[x2]=E [y2]-δ2 (2)
Step 3.3, the window of one (2R+1) × (2R+1) size is set, and coordinate is (i, j) at the window center position, is acquired
yi,jAverage value Qi,j
Wherein Qi,jTo pass through the pixel inner product and q of wavelet transformation in square windowi,jFor inner product and mean value, M is the size of shape window, R
For positive integer,
It enables
Step 3.4, according to formula (2) and?
Step 3.5, formula (5) is substituted into each coefficient that formula (1) obtains Wiener filtering
Step 3.6, according to constraints, formula (6) is converted to max (qi,j-δ2,0)；
Step 3.7, the least mean-square error for establishing Wiener filtering it is expected LMSE
LMSE=E { [xi,j-ai,jyi,j]2} (7)
Step 3.8, b is enabledi,j=δ2/qi,j, obtained by formula (6)
LSME=E { [xi,j-(1-bi,j)(xi,j-ni,j)]2} (8)
LSME=E { [bi,j(xi,j+ni,j)-ni,j]2} (9)
Wherein, b is independent from each other with x, n
Step 3.9, formula (5) formula (9) is substituted into obtain
Step 3.10, if
Then
Step 3.11, formula (12) is solved
qi,j< k δ2 (13)
Wherein,
Step 3.12, it if formula (13) is set up, abandons to xi,jLinear Estimation, x is assigned 0, is completed to the wavelet decomposition wiener of image
Algorithm；
Step 3.13, by all directions after the low frequency component of the wavelet transformation of noise-reduced image and after adaptive wiener filter point
Amount is overlapped, and reconstructs to obtain the image after noise reduction using db2 wavelet functions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710269122.4A CN108734669A (en) | 2017-04-24 | 2017-04-24 | Image denoising method based on wavelet transformation Wiener filtering and edge detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710269122.4A CN108734669A (en) | 2017-04-24 | 2017-04-24 | Image denoising method based on wavelet transformation Wiener filtering and edge detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108734669A true CN108734669A (en) | 2018-11-02 |
Family
ID=63934151
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710269122.4A Pending CN108734669A (en) | 2017-04-24 | 2017-04-24 | Image denoising method based on wavelet transformation Wiener filtering and edge detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108734669A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112927169A (en) * | 2021-04-06 | 2021-06-08 | 江苏海洋大学 | Remote sensing image denoising method based on wavelet transformation and improved weighted nuclear norm minimization |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102289821A (en) * | 2011-08-25 | 2011-12-21 | 西北工业大学 | Image detection method for side-slipping motion of vehicle |
| CN102496021A (en) * | 2011-11-23 | 2012-06-13 | 南开大学 | Wavelet transform-based thresholding method of image |
| CN102496153A (en) * | 2011-11-04 | 2012-06-13 | 西安电子科技大学 | SAR image speckle suppression method based on dictionary learning in wavelet domain |
| CN102521911A (en) * | 2011-12-16 | 2012-06-27 | 尤新革 | Identification method of crown word number (serial number) of bank note |
| CN102842134A (en) * | 2012-07-16 | 2012-12-26 | 西安电子科技大学 | Rapid scene matching method based on SAR (Synthetic Aperture Radar) image |
| CN103489157A (en) * | 2012-06-12 | 2014-01-01 | 中国科学院声学研究所 | Filtering method and system for enhancing synthetic aperture sonar interferogram quality |
| CN104504652A (en) * | 2014-10-10 | 2015-04-08 | 中国人民解放军理工大学 | Image denoising method capable of quickly and effectively retaining edge and directional characteristics |
| CN104778662A (en) * | 2014-12-25 | 2015-07-15 | 深圳市一体太赫兹科技有限公司 | Millimeter-wave image enhancing method and system |
-
2017
- 2017-04-24 CN CN201710269122.4A patent/CN108734669A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102289821A (en) * | 2011-08-25 | 2011-12-21 | 西北工业大学 | Image detection method for side-slipping motion of vehicle |
| CN102496153A (en) * | 2011-11-04 | 2012-06-13 | 西安电子科技大学 | SAR image speckle suppression method based on dictionary learning in wavelet domain |
| CN102496021A (en) * | 2011-11-23 | 2012-06-13 | 南开大学 | Wavelet transform-based thresholding method of image |
| CN102521911A (en) * | 2011-12-16 | 2012-06-27 | 尤新革 | Identification method of crown word number (serial number) of bank note |
| CN103489157A (en) * | 2012-06-12 | 2014-01-01 | 中国科学院声学研究所 | Filtering method and system for enhancing synthetic aperture sonar interferogram quality |
| CN102842134A (en) * | 2012-07-16 | 2012-12-26 | 西安电子科技大学 | Rapid scene matching method based on SAR (Synthetic Aperture Radar) image |
| CN104504652A (en) * | 2014-10-10 | 2015-04-08 | 中国人民解放军理工大学 | Image denoising method capable of quickly and effectively retaining edge and directional characteristics |
| CN104778662A (en) * | 2014-12-25 | 2015-07-15 | 深圳市一体太赫兹科技有限公司 | Millimeter-wave image enhancing method and system |
Non-Patent Citations (2)
| Title |
|---|
| 江陶: "基于 FPGA 小波域维纳滤波去噪算法的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 赵谦、侯媛彬、郑茂全: "《智能视频图像处理技术与应用》", 30 November 2016 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112927169A (en) * | 2021-04-06 | 2021-06-08 | 江苏海洋大学 | Remote sensing image denoising method based on wavelet transformation and improved weighted nuclear norm minimization |
| CN112927169B (en) * | 2021-04-06 | 2023-08-15 | 江苏海洋大学 | Remote sensing image denoising method based on wavelet transformation and improved weighted kernel norm minimization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Naimi et al. | Medical image denoising using dual tree complex thresholding wavelet transform and Wiener filter | |
| Knaus et al. | Dual-domain image denoising | |
| CN114140353A (en) | Swin-Transformer image denoising method and system based on channel attention | |
| CN108932699B (en) | Three-dimensional matching harmonic filtering image denoising method based on transform domain | |
| CN111815550B (en) | Infrared and visible light image fusion method based on gray level co-occurrence matrix | |
| KR20140109801A (en) | Method and apparatus for enhancing quality of 3D image | |
| KR102106073B1 (en) | Method and apparatus for processing images | |
| CN106651789B (en) | Self-adaptive deblocking method for compressed face image | |
| CN109993703A (en) | Multi-scale image noise-reduction method and device | |
| Rahman et al. | A new filtering technique for denoising speckle noise from medical images based on adaptive and anisotropic diffusion filter | |
| Sharumathi et al. | A survey on various image enhancement techniques for underwater acoustic images | |
| Anjana et al. | Color image enhancement using edge based histogram equalization | |
| Liu et al. | The Translation Invariant Wavelet-based Contourlet Transform for Image Denoising. | |
| Gao et al. | Infrared and visible image fusion using dual-tree complex wavelet transform and convolutional sparse representation | |
| CN110599406B (en) | Image enhancement method and device | |
| Hu et al. | A region-based video de-noising algorithm based on temporal and spatial correlations | |
| CN108734669A (en) | Image denoising method based on wavelet transformation Wiener filtering and edge detection | |
| Kumar et al. | Performance analysis of filters for speckle reduction in medical polycystic ovary ultrasound images | |
| Batna | Comparative study between wavelet thresholding techniques (hard, soft and invariant-translation) in ultrasound images | |
| CN113222833B (en) | Side-scan sonar image processing method and device | |
| CN113487505A (en) | Infrared image mixed noise reduction method based on noise identification | |
| Tun et al. | Joint Training of Noisy Image Patch and Impulse Response of Low-Pass Filter in CNN for Image Denoising | |
| CN111652808A (en) | Infrared image detail enhancement and noise suppression method | |
| Bhonslei | Additive and multiplicative noise removal from medical images using bivariate thresholding by dual tree complex wavelet transform | |
| Xuhui et al. | Ultrasound image denoising based on the contourlet transform and anisotropic diffusion |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181102 |