CN102968770A - Method and device for eliminating noise - Google Patents
Method and device for eliminating noise Download PDFInfo
- Publication number
- CN102968770A CN102968770A CN2012105042213A CN201210504221A CN102968770A CN 102968770 A CN102968770 A CN 102968770A CN 2012105042213 A CN2012105042213 A CN 2012105042213A CN 201210504221 A CN201210504221 A CN 201210504221A CN 102968770 A CN102968770 A CN 102968770A
- Authority
- CN
- China
- Prior art keywords
- noise
- pixel
- signal
- denoising
- gray
- 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
- 238000000034 method Methods 0.000 title claims abstract description 44
- 230000009466 transformation Effects 0.000 claims abstract description 55
- 230000006641 stabilisation Effects 0.000 claims abstract description 5
- 238000011105 stabilization Methods 0.000 claims abstract description 5
- 238000005070 sampling Methods 0.000 claims description 27
- 238000010586 diagram Methods 0.000 claims description 25
- 230000008030 elimination Effects 0.000 claims description 25
- 238000003379 elimination reaction Methods 0.000 claims description 25
- 238000006243 chemical reaction Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 description 35
- 239000000203 mixture Substances 0.000 description 9
- 230000001419 dependent effect Effects 0.000 description 7
- 238000012937 correction Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011524 similarity measure Methods 0.000 description 2
- 230000002146 bilateral effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- 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
- G06T2207/20076—Probabilistic image processing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the invention provides a method and device for eliminating noise. The method comprises the following steps of: obtaining a parameter estimated value of a noise standard deviation function of a first signal with noise to be eliminated based on a mixed noise model, so that an estimated standard deviation function is obtained; carrying out variance stabilization transformation on the first signal according to the estimated noise standard deviation function to obtain a second signal with noise being signal-uncorrelated noise; denoising the second signal; carrying out inverse transformation of the variance stabilization transformation on the denoised second signal, so that the noise of the first signal is eliminated. The embodiment of the invention provides the method and device for eliminating noise, so that mixed noise containing both a signal-correlated noise component and a signal-uncorrelated noise component is effectively eliminated.
Description
Technical field
The present invention relates to image processing techniques, relate in particular to a kind of noise cancellation method and device, belong to communication technical field.
Background technology
Image denoising is the hot research problem of image processing field always.
Traditional image de-noising method is mostly based on the thought of spatial domain part filter, such as mean filter, gaussian filtering, bilateral filtering etc.The spatial domain part filter is based on the pixel that closes on the locus and generally has this hypothesis of comparatively similar gray-scale value, by the gray-scale value of pixel and the gray-scale value of neighborhood territory pixel point are made weighted mean, removes noise.Because it is only applicable to the smooth of image that the pixel that the locus is closed on generally has the hypothesis of comparatively similar gray-scale value, and inapplicable to the detail section of image (edge, texture strong zone etc.), so the part filter method causes losing of image detail easily.
The image de-noising method that the people such as Buades proposed a kind of non-local mean in 2005.The non-local mean denoise algorithm is mainly utilized and is had a large amount of these redundant informations of self similarity piece in the digital picture, treat the similarity measure of the neighborhood of pixel points of denoising neighborhood of pixel points and region of search by foundation, calculate each pixel of region of search and the similarity weight for the treatment of the denoising pixel, then the pixel in the region of search is weighted on average, thereby calculates the gray-scale value for the treatment of that the denoising pixel is new.Although this algorithm has extraordinary effect in the maintenance of denoising performance and image texture, marginal information, it is based on the irrelevant Gaussian noise hypothesis of signal.
Existing noise-removed technology is mostly based on the irrelevant Gaussian noise model of signal, and in some practical application scenes, picture noise can be the mixed noise of Gaussian noise and signal dependent noise.For example, for the sensor imaging, the Gaussian noise component that the existing signal of noise model is irrelevant, the poisson noise component that signal correction is arranged again, therefore directly use the denoising method based on the Gaussian noise hypothesis, can't realize the effective denoising that comprises simultaneously the mixed noise of signal dependent noise component and signal uncorrelated noise component to this.
Summary of the invention
For the defective that exists in the prior art, the embodiment of the invention provides a kind of noise cancellation method and device, for the effective denoising that realizes the mixed noise that comprises simultaneously signal dependent noise component and signal uncorrelated noise component.
First aspect provides a kind of noise cancellation method, comprising:
Based on the mixed noise model, obtain the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, with the noise criteria difference function that obtains to estimate;
Noise criteria difference function according to estimating carries out variance-stabilizing transformation to described first signal, to obtain noise as the secondary signal of signal uncorrelated noise;
Described secondary signal is carried out denoising;
To the inverse transformation that the secondary signal after the denoising is carried out described variance-stabilizing transformation, finish the noise of described first signal is eliminated.
In the possible implementation of the first of first aspect, described variance-stabilizing transformation realizes by following formula:
Wherein,
Be the noise criteria difference function of described estimation, c is the constant standard deviation after the conversion, and t is current pixel gray-scale value before the conversion, f
VST(t) be current pixel gray-scale value after the conversion.
In the possible implementation of the first of first aspect, described based on the mixed noise model, obtain the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, the noise criteria difference function to obtain to estimate comprises:
Described first signal is carried out the wavelet field analysis, obtain (x, σ) scatter diagram;
Adopt random sampling consistency algorithm RANSAC, described (x, σ) scatter diagram is carried out curve fitting, obtain the first noise parameter a and the second noise parameter b, and:
Wherein, x is original noise-free signal corresponding to described first signal,
Noise criteria difference function for described estimation.
In conjunction with the possible implementation of first or the second of first aspect or first aspect, in the third possible implementation of first aspect, described described secondary signal is carried out denoising, comprise each pixel that travels through in such a way described secondary signal:
Judge the neighborhood of the pixel i treat denoising and the ratio of the gray average of the neighborhood of the pixel j of region of search, with 1 difference whether less than or equal to preset difference value; And whether the angle of gradient direction of judging described pixel i and described pixel j is less than or equal to default angle; Wherein i and j are natural number;
If at least one among both is judged as no, then the neighborhood similarity with described pixel i and described pixel j is defined as 0;
If both all are judged as, then according to default formula, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
In conjunction with the possible implementation of first or the second of first aspect or first aspect, in the 4th kind of possible implementation of first aspect, described described secondary signal is carried out denoising, comprise each pixel that travels through in such a way described secondary signal:
The neighborhood window for the treatment of the pixel j of the pixel i of denoising and region of search carries out down-sampling; Wherein i and j are natural number;
According to the gray-scale value of the down-sampling neighborhood of described pixel i, with the gray-scale value of the down-sampling neighborhood of described pixel j, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
Second aspect provides a kind of noise elimination apparatus, comprising:
Estimation module is used for based on the mixed noise model, obtains the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, with the noise criteria difference function that obtains to estimate;
The variance-stabilizing transformation module is used for according to the noise criteria difference function of estimating described first signal being carried out variance-stabilizing transformation, to obtain noise as the secondary signal of signal uncorrelated noise;
The denoising module is used for described secondary signal is carried out denoising;
Variance stabilization inverse transform block to the inverse transformation that the secondary signal after the denoising is carried out described variance-stabilizing transformation, is finished the noise of described first signal is eliminated.
In the possible implementation of the first of second aspect, described variance-stabilizing transformation realizes by following formula:
Wherein,
Be the noise criteria difference function of described estimation, c is the constant standard deviation after the conversion, and t is current pixel gray-scale value before the conversion, f
VST(t) be current pixel gray-scale value after the conversion.
In the possible implementation of the second of second aspect, described estimation module is used for:
First signal is carried out the wavelet field analysis, obtain (x, σ) scatter diagram;
Adopt random sampling consistency algorithm RANSAC, described (x, σ) scatter diagram is carried out curve fitting, obtain the first noise parameter a and the second noise parameter b, and:
Wherein, x is original noise-free signal corresponding to described first signal,
Noise criteria difference function for described estimation.
In conjunction with the possible implementation of first or the second of second aspect or second aspect, in the third possible implementation of second aspect, described denoising module is used for traveling through in such a way each pixel of described secondary signal:
Judge the neighborhood of the pixel i treat denoising and the ratio of the gray average of the neighborhood of the pixel j of region of search, with 1 difference whether less than or equal to preset difference value; And whether the angle of gradient direction of judging described pixel i and described pixel j is less than or equal to default angle; Wherein i and j are natural number;
If at least one among both is judged as no, then the neighborhood similarity with described pixel i and described pixel j is defined as 0;
If both all are judged as, then according to default formula, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
In conjunction with the possible implementation of first or the second of second aspect or second aspect, in the 4th kind of possible implementation of second aspect, described denoising module is used for traveling through in such a way each pixel of described secondary signal:
The neighborhood window for the treatment of the pixel j of the pixel i of denoising and region of search carries out down-sampling; Wherein i and j are natural number;
According to the gray-scale value of the down-sampling neighborhood of described pixel i, with the gray-scale value of the down-sampling neighborhood of described pixel j, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
The noise cancellation method and the device that provide according to the embodiment of the invention, by based on the mixed noise model, signal with mixed noise is carried out the noise criteria difference function to be estimated, and utilize the noise criteria difference function to carry out variance-stabilizing transformation, the signal of noise to be eliminated is converted to the signal with signal uncorrelated noise, carries out denoising thereby can utilize arbitrarily based on the denoising method of signal uncorrelated noise hypothesis.Therefore, realized effective elimination for the mixed noise that comprises simultaneously signal dependent noise component and signal uncorrelated noise component.
Description of drawings
In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art, the below will do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art, apparently, accompanying drawing in the following describes only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain according to these accompanying drawings other accompanying drawing.
Fig. 1 is the schematic flow sheet of the noise cancellation method of the embodiment of the invention.
Fig. 2 is for carrying out the principle schematic that noise parameter is estimated based on the wavelet field analysis.
Fig. 3 is an example of (x, σ) scatter diagram.
Fig. 4 is the result schematic diagram that directly (x, σ) scatter diagram shown in Figure 3 is carried out curve fitting.
The result schematic diagram of Fig. 5 for adopting RANSAC that (x, σ) scatter diagram is carried out curve fitting.
Fig. 6 is the structural representation of the noise elimination apparatus of one embodiment of the invention.
Fig. 7 is the structural representation of the noise elimination apparatus of another embodiment of the present invention.
Embodiment
Fig. 1 is the schematic flow sheet of the noise cancellation method of the embodiment of the invention.As shown in Figure 1, this noise cancellation method comprises:
101, based on the mixed noise model, obtain the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, with the noise criteria difference function that obtains to estimate;
102, the noise criteria difference function according to estimating carries out variance-stabilizing transformation to described first signal, to obtain noise as the secondary signal of signal uncorrelated noise;
103, described secondary signal is carried out denoising;
104, to the inverse transformation that the secondary signal after the denoising is carried out described variance-stabilizing transformation, finish the noise of described first signal is eliminated.
Hereinafter, take the mixed noise model of sensor imaging as example, 101-104 is elaborated to above-mentioned steps.
Particularly, the mixed noise model of sensor imaging is Poisson-Gaussian Mixture noise model, shown in following formula (1):
Y=x+v
P(x)+v
GFormula (1)
Wherein, y is the signal (being first signal) that contains mixed noise, for example is the image of sensor imaging, and x is original noise-free signal, v
P(x) be the poisson noise composition of signal correction, v
GIt is the irrelevant Gaussian noise composition of signal.
For the poisson noise composition, satisfy following formula (2) and (3):
Formula (2)
Var (v
P)=ax formula (3)
Wherein, Var (v
P) be the noise variance of poisson noise composition.
For the Gaussian noise composition, satisfy following formula (4) and (5):
Var (v
G)=b formula (5)
Wherein, Var (v
G) be the noise variance of Gaussian noise composition.
Therefore, the noise variance of mixed noise satisfies following formula (6):
σ
2=ax+b formula (6)
Wherein, σ
2Noise variance for mixed noise; A and b are noise parameter, and it can adopt any-mode to realize, for example carry out noise parameter based on the wavelet field analysis and estimate.
Fig. 2 is for carrying out the principle schematic that noise parameter is estimated based on the wavelet field analysis.As shown in Figure 2, image y is carried out two-dimensional discrete wavelet conversion, obtain W
A, W
H, W
VAnd W
DFour different wavelet coefficient subgraphs, wherein W
ABe approximation coefficient, W
HBe level detail, W
VBe vertical detail, W
DBe the diagonal line details; With W
AThe gradient of each pixel is compared with default Grad in the subgraph, removes W
AThe larger pixel of gradient stays smooth region in the subgraph; The pixel that smooth region is comprised is divided into N level set S according to gray scale
1, S
2..., S
N, tonal range corresponding to different level set wherein; At each level set S
iUtilize W in (i=[1, N])
AEstimate x
i, utilize W
DEstimate σ
iAnd the method for use curve match, (x, the σ) scatter diagram that obtains is carried out match, obtain noise parameter a and b.
In addition, because for fixation of sensor, noise parameter a and b fix, therefore can be according to some images of sensor shooting, noise parameter a and b are estimated, and will estimate that the noise parameter a and the b that obtain store, when carrying out the successive image denoising, directly to call.
According to the noise parameter a and the b that estimate to obtain, can obtain the noise criteria difference function of the estimation shown in following formula (7):
After obtaining the noise criteria difference function, image y is carried out variance-stabilizing transformation (VarianceStabilizing Transformation, VST).Wherein, the effect of variance-stabilizing transformation is the Nonlinear Mapping by gray-scale value, and the noise that signal correction, variance are changed changes the noise that signal is irrelevant, variance is constant into, and the noise of the image after the conversion (being secondary signal) can be thought Gaussian noise.
Variance-stabilizing transformation for example realizes by following formula (8):
Wherein,
Be the noise criteria difference function of estimating; C is the constant standard deviation after the conversion, and it can be any number greater than 0, for example is set to 0.01; T is current pixel point, carries out the front gray-scale value of variance-stabilizing transformation; f
VST(t) be current pixel point, the gray-scale value after the execution variance-stabilizing transformation.
After carrying out variance-stabilizing transformation, can use based on any denoising method of Gaussian noise and carry out denoising, such as the image de-noising method that adopts non-local mean etc.
After finishing denoising, re-use the inverse transformation of variance-stabilizing transformation, gray-scale value is shone upon, thereby obtain the image behind the image y elimination noise so far, is finished the denoising to the image y with mixed noise.
Wherein, the relation of the inverse transformation of variance-stabilizing transformation and variance-stabilizing transformation satisfies following formula (9):
For Poisson-Gaussian Mixture noise model, the expression formula of the inverse transformation of variance-stabilizing transformation is for example shown in following formula (10) and (11):
Need to prove: in said process, although take the Poisson of sensor imaging-Gaussian Mixture noise as example, detailed process to the noise cancellation method of above-described embodiment is illustrated, but those skilled in the art can understand, for satisfying arbitrarily the mixed noise that noise variance is the monotonic quantity of signal intensity, all can carry out by the noise cancellation method of above-described embodiment noise and eliminate.
In addition, the specific implementation of variance-stabilizing transformation also can be taked the alternate manner different from formula (8), as long as the signal correction variance can be become the irrelevant variance of signal, do not limit in the embodiment of the invention, for example adopt following formula (12) to realize variance-stabilizing transformation:
Noise cancellation method according to above-described embodiment, by based on the mixed noise model, signal with mixed noise is carried out the noise criteria difference function to be estimated, and utilize the noise criteria difference function to carry out variance-stabilizing transformation, the signal of noise to be eliminated is converted to the signal with signal uncorrelated noise, carries out denoising thereby can utilize arbitrarily based on the denoising method of signal uncorrelated noise hypothesis.Therefore, realized effective elimination for the mixed noise that comprises simultaneously signal dependent noise component and signal uncorrelated noise component.
Further, on the basis of the noise cancellation method of above-described embodiment, the process of the estimates of parameters of the noise criteria difference function of the first signal that obtains noise to be eliminated is optimized.Particularly, carry out noise parameter based on the wavelet field analysis and estimate, get access to (x, σ) scatter diagram after, adopt random sampling consistency algorithm (RANSAC), (x, σ) scatter diagram is carried out curve fitting, to obtain noise parameter a and b.
More specifically, adopting RANSAC that (x, σ) scatter diagram is carried out curve fitting may further comprise the steps:
Whether the noise model curve of step 2, the match of inspection institute satisfies a 〉=0 and b 〉=0, if do not satisfy repeated execution of steps 1;
Step 3, have a few, statistics is from the distance of the matched curve number less than the point (hereinafter this point being called interior point) of given threshold value.If in count above previous iteration obtain in count, then be recorded as imperial palace and count.
Step 4, count according to imperial palace and to calculate the iterations of needs.
Step 5, iterative step 1 ~ 4 are until iterations is enough.
Point re-starts models fitting in the noise model curve of counting maximum in step 6, the usefulness all.
Fig. 3 is an example of (x, σ) scatter diagram; Fig. 4 is the result schematic diagram that directly (x, σ) scatter diagram shown in Figure 3 is carried out curve fitting; The result schematic diagram of Fig. 5 for adopting RANSAC that (x, σ) scatter diagram is carried out curve fitting.Can find out according to Fig. 3-5, by adopting RANSAC (x, σ) scatter diagram be carried out curve fitting, significantly improve the robustness that noise parameter is estimated, thus Effective Raise noise removing performance.
Further, on the basis of the noise cancellation method of above-described embodiment, the process of carrying out denoising to carrying out signal (being secondary signal) after the variance-stabilizing transformation is optimized.
Particularly, the image de-noising method that utilizes existing non-local mean is to carrying out the signal after the variance-stabilizing transformation, when carrying out denoising, it is the similarity measure for the treatment of the pixel neighborhood of a point of denoising pixel neighborhood of a point and region of search by foundation, calculate each pixel of region of search and the similarity weight for the treatment of the denoising pixel, then the pixel in the region of search is weighted on average, thereby calculates the gray-scale value for the treatment of that the denoising pixel is new, be i.e. gray-scale value after the denoising.
Gray-scale value after the denoising of denoising pixel calculates by following formula (12):
Wherein, pixel i is the pixel for the treatment of denoising; Pixel j is the pixel in the region of search; I is whole pixels of the image after the execution variance-stabilizing transformation; V (j) is the gray-scale value of pixel j; NL (v) is gray-scale value after the pixel i denoising (i); W (i, j) is the neighborhood similarity of the pixel j in J the pixel of pixel i and region of search; W (i, j) calculates by following formula (13):
Wherein, N
iNeighborhood for pixel i; N
jNeighborhood for pixel j; H is filtering severity control parameter, is generally determined by noise variance; Z (i) is normaliztion constant, satisfies following formula (14):
In the embodiment of the invention, provide following two kinds to the above-mentioned scheme that is optimized based on the image de-noising method of non-local mean:
Scheme one: the number of times that calculates the neighborhood similarity is optimized;
Judge the neighborhood of the pixel i treat denoising and the ratio of the gray average of the neighborhood of the pixel j of region of search, with 1 difference whether less than or equal to preset difference value; And whether the angle of gradient direction of judging pixel i and pixel j is less than or equal to default angle;
If at least one among both is judged as no, then the neighborhood similarity with pixel i and pixel j is defined as 0;
If both all are judged as, then according to default formula, the neighborhood similarity of calculating pixel point i and pixel j;
According to the gray-scale value of described neighborhood similarity and pixel j, calculate the gray-scale value after the denoising of described pixel i.
Particularly, in scheme one, the neighborhood of definition pixel i and the ratio of the gray average of the neighborhood of pixel j, such as following formula (15) institute formula:
Wherein, N
iNeighborhood for pixel i; N
jNeighborhood for pixel j; M (N
i) be the gray average of the neighborhood of pixel i; M (N
j) be the gray average of the neighborhood of pixel j; R (i, j) is N
iWith N
jThe ratio of gray average;
Also define the angle of the gradient direction of pixel i and pixel j, such as following formula (16) institute formula:
Wherein, G (i) is the gradient direction of pixel i; G (j) is the gradient direction of pixel j; θ (i, j) is the angle of the gradient direction of pixel i and pixel j.
Gray average for neighborhood differs larger, and perhaps the pixel that differs greatly of gradient direction does not calculate its neighborhood similarity, directly thinks 0.Based on this thought, in conjunction with above-mentioned definition, be following formula (17) with the neighborhood calculating formula of similarity abbreviation shown in the formula (13):
Wherein, η
1, η
2Be the numerical value that sets in advance as required, wherein η with ζ
1For less than 1 positive number, for example be set to 0.9; η
2For greater than 1 numerical value, for example be set to 1.1; ζ is the angle value less than or equal to 90 degree, for example is set to 60 degree.
By such scheme one, can effectively reduce the number of times that calculates the neighborhood similarity.
Scheme two: the complexity of calculating the neighborhood similarity is optimized;
The neighborhood window for the treatment of the pixel j of the pixel i of denoising and region of search carries out down-sampling;
According to the gray-scale value of the down-sampling neighborhood of pixel i, and the gray-scale value of the down-sampling neighborhood of pixel j, the neighborhood similarity of calculating pixel point i and pixel j.
Particularly, in scheme two, utilize the down-sampling of neighborhood window to carry out calculating.Because the continuity of image, the Weighted distance of down-sampling neighborhood is similar to the Weighted distance of former neighborhood.
Based on this thought, be following formula (18) with the neighborhood calculating formula of similarity abbreviation shown in the formula (13):
By such scheme two, can significantly reduce the complexity of the calculating of neighborhood similarity.
Such scheme one and scheme two both can be used separately, also can be combined with, and limited in the embodiment of the invention.
Fig. 6 is the structural representation of the noise elimination apparatus of one embodiment of the invention.As shown in Figure 6, this noise elimination apparatus 60 comprises:
Variance-stabilizing transformation module 62 is used for according to the noise criteria difference function of estimating described first signal being carried out variance-stabilizing transformation, to obtain noise as the secondary signal of signal uncorrelated noise;
Variance stabilization inverse transform block 64 to the inverse transformation that the secondary signal after the denoising is carried out described variance-stabilizing transformation, is finished the noise of described first signal is eliminated.
The idiographic flow that the noise elimination apparatus of above-described embodiment is eliminated noise is identical with the noise cancellation method of above-described embodiment, so locate to repeat no more.
Noise elimination apparatus according to above-described embodiment, by based on the mixed noise model, signal with mixed noise is carried out the noise criteria difference function to be estimated, and utilize the noise criteria difference function to carry out variance-stabilizing transformation, the signal of noise to be eliminated is converted to the signal with signal uncorrelated noise, carries out denoising thereby can utilize arbitrarily based on the denoising method of signal uncorrelated noise hypothesis.Therefore, realized effective elimination for the mixed noise that comprises simultaneously signal dependent noise component and signal uncorrelated noise component.
Further, in the noise elimination apparatus of above-described embodiment, described variance-stabilizing transformation realizes by following formula:
Wherein,
Be the noise criteria difference function of described estimation, c is the constant standard deviation after the conversion, and t is current pixel gray-scale value before the conversion, f
VST(t) be current pixel gray-scale value after the conversion.
Further, in the noise elimination apparatus of above-described embodiment, described estimation module is used for:
First signal is carried out the wavelet field analysis, obtain (x, σ) scatter diagram;
Adopt random sampling consistency algorithm RANSAC, described (x, σ) scatter diagram is carried out curve fitting, obtain the first noise parameter a and the second noise parameter b, and:
Wherein, x is original noise-free signal corresponding to described first signal,
Noise criteria difference function for described estimation.
Further, in the noise elimination apparatus of above-described embodiment, described denoising module is used for traveling through in such a way each pixel of described secondary signal:
Judge the neighborhood of the pixel i treat denoising and the ratio of the gray average of the neighborhood of the pixel j of region of search, with 1 difference whether less than or equal to preset difference value; And whether the angle of gradient direction of judging pixel i and pixel j is less than or equal to default angle; Wherein i and j are natural number;
If at least one among both is judged as no, then the neighborhood similarity with pixel i and pixel j is defined as 0;
If both all are judged as, then according to default formula, the neighborhood similarity of calculating pixel point i and pixel j;
According to the gray-scale value of described neighborhood similarity and pixel j, calculate the gray-scale value after the denoising of described pixel i.
Further, in the noise elimination apparatus of above-described embodiment, described denoising module is used for traveling through in such a way each pixel of described secondary signal:
The neighborhood window for the treatment of the pixel j of the pixel i of denoising and region of search carries out down-sampling; Wherein i and j are natural number;
According to the gray-scale value of the down-sampling neighborhood of pixel i, and the gray-scale value of the down-sampling neighborhood of pixel j, the neighborhood similarity of calculating pixel point i and pixel j;
According to the gray-scale value of described neighborhood similarity and pixel j, calculate the gray-scale value after the denoising of described pixel i.
Fig. 7 is the structural representation of the noise elimination apparatus of another embodiment of the present invention.As shown in Figure 7, this noise elimination apparatus 70 comprises storer 71 and processor 72, wherein:
Storage batch processing code in the storer 71, and processor 72 is used for carrying out following the operation for the program code that calls storer 71 storages:
Based on the mixed noise model, obtain the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, with the noise criteria difference function that obtains to estimate;
Noise criteria difference function according to estimating carries out variance-stabilizing transformation to described first signal, to obtain noise as the secondary signal of signal uncorrelated noise;
Described secondary signal is carried out denoising;
To the inverse transformation that the secondary signal after the denoising is carried out described variance-stabilizing transformation, finish the noise of described first signal is eliminated.
The idiographic flow that the noise elimination apparatus of above-described embodiment is eliminated noise is identical with the noise cancellation method of above-described embodiment, so locate to repeat no more.
Noise elimination apparatus according to above-described embodiment, by based on the mixed noise model, signal with mixed noise is carried out the noise criteria difference function to be estimated, and utilize the noise criteria difference function to carry out variance-stabilizing transformation, the signal of noise to be eliminated is converted to the signal with signal uncorrelated noise, carries out denoising thereby can utilize arbitrarily based on the denoising method of signal uncorrelated noise hypothesis.Therefore, realized effective elimination for the mixed noise that comprises simultaneously signal dependent noise component and signal uncorrelated noise component.
One of ordinary skill in the art will appreciate that: all or part of step that realizes above-mentioned each embodiment of the method can be finished by the relevant hardware of programmed instruction.Aforesaid program can be stored in the computer read/write memory medium.This program is carried out the step that comprises above-mentioned each embodiment of the method when carrying out; And aforesaid storage medium comprises: the various media that can be program code stored such as ROM, RAM, magnetic disc or CD.
It should be noted that at last: above embodiment only in order to technical scheme of the present invention to be described, is not intended to limit; Although with reference to previous embodiment the present invention is had been described in detail, those of ordinary skill in the art is to be understood that: it still can be made amendment to the technical scheme that aforementioned each embodiment puts down in writing, and perhaps part technical characterictic wherein is equal to replacement; And these modifications or replacement do not make the essence of appropriate technical solution break away from the spirit and scope of various embodiments of the present invention technical scheme.
Claims (10)
1. a noise cancellation method is characterized in that, comprising:
Based on the mixed noise model, obtain the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, with the noise criteria difference function that obtains to estimate;
Noise criteria difference function according to estimating carries out variance-stabilizing transformation to described first signal, to obtain noise as the secondary signal of signal uncorrelated noise;
Described secondary signal is carried out denoising;
To the inverse transformation that the secondary signal after the denoising is carried out described variance-stabilizing transformation, finish the noise of described first signal is eliminated.
2. noise cancellation method according to claim 1 is characterized in that, described variance-stabilizing transformation realizes by following formula:
3. noise cancellation method according to claim 1 is characterized in that, and is described based on the mixed noise model, obtains the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, and the noise criteria difference function to obtain to estimate comprises:
Described first signal is carried out the wavelet field analysis, obtain (x, σ) scatter diagram;
Adopt random sampling consistency algorithm RANSAC, described (x, σ) scatter diagram is carried out curve fitting, obtain the first noise parameter a and the second noise parameter b, and:
4. arbitrary described noise cancellation method is characterized in that according to claim 1-3, described described secondary signal is carried out denoising, comprises each pixel that travels through in such a way described secondary signal:
Judge the neighborhood of the pixel i treat denoising and the ratio of the gray average of the neighborhood of the pixel j of region of search, with 1 difference whether less than or equal to preset difference value; And whether the angle of gradient direction of judging described pixel i and described pixel j is less than or equal to default angle; Wherein i and j are natural number;
If at least one among both is judged as no, then the neighborhood similarity with described pixel i and described pixel j is defined as 0;
If both all are judged as, then according to default formula, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
5. arbitrary described noise cancellation method is characterized in that according to claim 1-3, described described secondary signal is carried out denoising, comprises each pixel that travels through in such a way described secondary signal:
The neighborhood window for the treatment of the pixel j of the pixel i of denoising and region of search carries out down-sampling; Wherein i and j are natural number;
According to the gray-scale value of the down-sampling neighborhood of described pixel i, with the gray-scale value of the down-sampling neighborhood of described pixel j, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
6. a noise elimination apparatus is characterized in that, comprising:
Estimation module is used for based on the mixed noise model, obtains the estimates of parameters of noise criteria difference function of the first signal of noise to be eliminated, with the noise criteria difference function that obtains to estimate;
The variance-stabilizing transformation module is used for according to the noise criteria difference function of estimating described first signal being carried out variance-stabilizing transformation, to obtain noise as the secondary signal of signal uncorrelated noise;
The denoising module is used for described secondary signal is carried out denoising;
Variance stabilization inverse transform block to the inverse transformation that the secondary signal after the denoising is carried out described variance-stabilizing transformation, is finished the noise of described first signal is eliminated.
7. noise elimination apparatus according to claim 6 is characterized in that, described variance-stabilizing transformation realizes by following formula:
8. noise elimination apparatus according to claim 6 is characterized in that, described estimation module is used for:
First signal is carried out the wavelet field analysis, obtain (x, σ) scatter diagram;
Adopt random sampling consistency algorithm RANSAC, described (x, σ) scatter diagram is carried out curve fitting, obtain the first noise parameter a and the second noise parameter b, and:
9. arbitrary described noise elimination apparatus is characterized in that according to claim 6-8, and described denoising module is used for traveling through in such a way each pixel of described secondary signal:
Judge the neighborhood of the pixel i treat denoising and the ratio of the gray average of the neighborhood of the pixel j of region of search, with 1 difference whether less than or equal to preset difference value; And whether the angle of gradient direction of judging described pixel i and described pixel j is less than or equal to default angle; Wherein i and j are natural number;
If at least one among both is judged as no, then the neighborhood similarity with described pixel i and described pixel j is defined as 0;
If both all are judged as, then according to default formula, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
10. arbitrary described noise elimination apparatus is characterized in that according to claim 6-8, and described denoising module is used for traveling through in such a way each pixel of described secondary signal:
The neighborhood window for the treatment of the pixel j of the pixel i of denoising and region of search carries out down-sampling; Wherein i and j are natural number;
According to the gray-scale value of the down-sampling neighborhood of described pixel i, with the gray-scale value of the down-sampling neighborhood of described pixel j, calculate the neighborhood similarity of described pixel i and described pixel j;
According to the gray-scale value of described neighborhood similarity and described pixel j, calculate the gray-scale value after the denoising of described pixel i.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012105042213A CN102968770A (en) | 2012-11-30 | 2012-11-30 | Method and device for eliminating noise |
PCT/CN2013/077669 WO2014082441A1 (en) | 2012-11-30 | 2013-06-21 | Noise elimination method and apparatus |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012105042213A CN102968770A (en) | 2012-11-30 | 2012-11-30 | Method and device for eliminating noise |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102968770A true CN102968770A (en) | 2013-03-13 |
Family
ID=47798898
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2012105042213A Pending CN102968770A (en) | 2012-11-30 | 2012-11-30 | Method and device for eliminating noise |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN102968770A (en) |
WO (1) | WO2014082441A1 (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014082441A1 (en) * | 2012-11-30 | 2014-06-05 | 华为技术有限公司 | Noise elimination method and apparatus |
CN108171118A (en) * | 2017-12-05 | 2018-06-15 | 东软集团股份有限公司 | Signal of blinking data processing method, device, readable storage medium storing program for executing and electronic equipment |
CN108899041A (en) * | 2018-08-20 | 2018-11-27 | 百度在线网络技术(北京)有限公司 | Voice signal adds method for de-noising, device and storage medium |
CN109345489A (en) * | 2018-10-31 | 2019-02-15 | 福州大学 | Error diffusion method based on image pixel neighborhood grayscale information |
CN109559281A (en) * | 2017-09-26 | 2019-04-02 | 三星电子株式会社 | Image denoising neural network framework and its training method |
CN110730280A (en) * | 2018-07-17 | 2020-01-24 | 瑞昱半导体股份有限公司 | Noise equalization method and noise removal method |
CN111667428A (en) * | 2020-06-05 | 2020-09-15 | 北京百度网讯科技有限公司 | Noise generation method and device based on automatic search |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111784733B (en) * | 2020-07-06 | 2024-04-16 | 深圳市安健科技股份有限公司 | Image processing method, device, terminal and computer readable storage medium |
CN113129235A (en) * | 2021-04-22 | 2021-07-16 | 深圳市深图医学影像设备有限公司 | Medical image noise suppression algorithm |
CN113598786A (en) * | 2021-09-03 | 2021-11-05 | 青岛雅斯生物科技有限公司 | Electrocardio monitoring devices |
CN117408890B (en) * | 2023-12-14 | 2024-03-08 | 武汉泽塔云科技股份有限公司 | Video image transmission quality enhancement method and system |
CN117788570B (en) * | 2024-02-26 | 2024-05-07 | 山东济矿鲁能煤电股份有限公司阳城煤矿 | Bucket wheel machine positioning method and system based on machine vision |
CN117911792B (en) * | 2024-03-15 | 2024-06-04 | 垣矽技术(青岛)有限公司 | Pin detecting system for voltage reference source chip production |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1836266A (en) * | 2003-09-24 | 2006-09-20 | 爱信精机株式会社 | Device for detecting road traveling lane |
CN1836267A (en) * | 2003-09-24 | 2006-09-20 | 爱信精机株式会社 | Device for detecting road traveling lane |
CN102663396A (en) * | 2012-03-09 | 2012-09-12 | 华中科技大学 | Method for automatically detecting rice milky ripe stage |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5923775A (en) * | 1996-04-04 | 1999-07-13 | Eastman Kodak Company | Apparatus and method for signal dependent noise estimation and reduction in digital images |
US7260272B2 (en) * | 2003-07-10 | 2007-08-21 | Samsung Electronics Co.. Ltd. | Method and apparatus for noise reduction using discrete wavelet transform |
CN101582984B (en) * | 2009-04-14 | 2014-04-16 | 公安部物证鉴定中心 | Method and device for eliminating image noise |
CN102117482B (en) * | 2011-04-13 | 2013-04-17 | 西安电子科技大学 | Non-local mean image denoising method combined with structure information |
CN102968770A (en) * | 2012-11-30 | 2013-03-13 | 华为技术有限公司 | Method and device for eliminating noise |
-
2012
- 2012-11-30 CN CN2012105042213A patent/CN102968770A/en active Pending
-
2013
- 2013-06-21 WO PCT/CN2013/077669 patent/WO2014082441A1/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1836266A (en) * | 2003-09-24 | 2006-09-20 | 爱信精机株式会社 | Device for detecting road traveling lane |
CN1836267A (en) * | 2003-09-24 | 2006-09-20 | 爱信精机株式会社 | Device for detecting road traveling lane |
CN102663396A (en) * | 2012-03-09 | 2012-09-12 | 华中科技大学 | Method for automatically detecting rice milky ripe stage |
Non-Patent Citations (4)
Title |
---|
CHAO PANG等: "《Multimedia Signal Processing, 2009, MMSP "09. IEEE International Workshop on》", 7 October 2009 * |
MAHMOUDI, M等: "《Fast image and video denoising via nonlocal means of similar neighborhoods》", 《SIGNAL PROCESSING LETTERS, IEEE》 * |
MINCHAO YE等: "《Hyperspectral Image and Signal Processing (WHISPERS), 2012 4th Workshop on》", 7 June 2012 * |
钟锦敏: "《非局部平均的去噪方法研究》", 《万方学位论文》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014082441A1 (en) * | 2012-11-30 | 2014-06-05 | 华为技术有限公司 | Noise elimination method and apparatus |
CN109559281A (en) * | 2017-09-26 | 2019-04-02 | 三星电子株式会社 | Image denoising neural network framework and its training method |
CN109559281B (en) * | 2017-09-26 | 2023-12-08 | 三星电子株式会社 | Image denoising neural network architecture and training method thereof |
CN108171118A (en) * | 2017-12-05 | 2018-06-15 | 东软集团股份有限公司 | Signal of blinking data processing method, device, readable storage medium storing program for executing and electronic equipment |
CN108171118B (en) * | 2017-12-05 | 2020-10-02 | 东软集团股份有限公司 | Blink signal data processing method and device, readable storage medium and electronic equipment |
CN110730280A (en) * | 2018-07-17 | 2020-01-24 | 瑞昱半导体股份有限公司 | Noise equalization method and noise removal method |
CN110730280B (en) * | 2018-07-17 | 2021-08-31 | 瑞昱半导体股份有限公司 | Noise equalization method and noise removal method |
CN108899041A (en) * | 2018-08-20 | 2018-11-27 | 百度在线网络技术(北京)有限公司 | Voice signal adds method for de-noising, device and storage medium |
CN108899041B (en) * | 2018-08-20 | 2019-12-27 | 百度在线网络技术(北京)有限公司 | Voice signal noise adding method, device and storage medium |
CN109345489A (en) * | 2018-10-31 | 2019-02-15 | 福州大学 | Error diffusion method based on image pixel neighborhood grayscale information |
CN111667428A (en) * | 2020-06-05 | 2020-09-15 | 北京百度网讯科技有限公司 | Noise generation method and device based on automatic search |
Also Published As
Publication number | Publication date |
---|---|
WO2014082441A1 (en) | 2014-06-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102968770A (en) | Method and device for eliminating noise | |
Wang et al. | Dehazing for images with large sky region | |
Chen et al. | Spatially adapted total variation model to remove multiplicative noise | |
US20160321523A1 (en) | Using machine learning to filter monte carlo noise from images | |
Rabie | Robust estimation approach for blind denoising | |
CN105335947A (en) | Image de-noising method and image de-noising apparatus | |
CN106127688B (en) | A kind of super-resolution image reconstruction method and its system | |
CN105913396A (en) | Noise estimation-based image edge preservation mixed de-noising method | |
CN103942758A (en) | Dark channel prior image dehazing method based on multiscale fusion | |
CN101882304A (en) | Self-adaptive de-noising and characteristic enhancing method of SAR (Synthetic Aperture Radar) image | |
CN101901476A (en) | SAR image de-noising method based on NSCT domain edge detection and Bishrink model | |
EP3105736A1 (en) | Method for performing super-resolution on single images and apparatus for performing super-resolution on single images | |
CN102073994B (en) | Ultrasonic medical image speckle noise inhibition method based on multi-scale anisotropic diffusion | |
CN104992421A (en) | OpenCL-based parallel optimization method of image de-noising algorithm | |
CN110349112A (en) | A kind of two stages image de-noising method based on adaptive singular value threshold value | |
CN102789634B (en) | A kind of method obtaining illumination homogenization image | |
CN104200434A (en) | Non-local mean image denoising method based on noise variance estimation | |
CN104616259A (en) | Non-local mean image de-noising method with noise intensity self-adaptation function | |
CN115082336A (en) | SAR image speckle suppression method based on machine learning | |
CN111091107A (en) | Face region edge detection method and device and storage medium | |
CN111652821B (en) | Low-light video image noise reduction processing method, device and equipment based on gradient information | |
CN105046665B (en) | A kind of Wavelet noise-eliminating method for high throughput gene sequencing image | |
CN106934398A (en) | Image de-noising method based on super-pixel cluster and rarefaction representation | |
Karthikeyan et al. | Energy based denoising convolutional neural network for image enhancement | |
CN117830134A (en) | Infrared image enhancement method and system based on mixed filtering decomposition and image fusion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130313 |