CN106464778B - The method and device of noise suppressed is carried out to image - Google Patents
The method and device of noise suppressed is carried out to image Download PDFInfo
- Publication number
- CN106464778B CN106464778B CN201480077057.XA CN201480077057A CN106464778B CN 106464778 B CN106464778 B CN 106464778B CN 201480077057 A CN201480077057 A CN 201480077057A CN 106464778 B CN106464778 B CN 106464778B
- Authority
- CN
- China
- Prior art keywords
- image
- noise
- constraint condition
- periodic
- periodic noise
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N5/00—Details of television systems
- H04N5/14—Picture signal circuitry for video frequency region
- H04N5/21—Circuitry for suppressing or minimising disturbance, e.g. moiré or halo
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
Abstract
A kind of pair of image carries out the method and device of noise suppressed, and method includes: to obtain image to be processed, and image to be processed is the image interfered by periodic noise；Determine the period of periodic noise；The period of the periodic noise is substituted into signals separating model, signals separating model is optimized, obtains the clean picture signal corresponding to the image to be processed.Based on SPX model realization, the period that the periodic noise determined is combined in the model is handled, so that when isolating periodic noise signal, aperiodicity noise signal and clean picture signal, it is contemplated that the cyclic effects of noise, so that not including ringing effect in clean image, and then improve picture quality.
Description
Technical field
It is special this application involves the method and device that image processing techniques more particularly to a kind of pair of image carry out noise suppressed
Not, which is the image interfered by periodic noise.
Background technique
Digital radiographic art (Digital Radiography, DR) be it is a kind of using digital sensor detect X-ray into
Row shoots the technology of medicine photograph, stores and processs medical image in a manner of number, has acquisition and transmission speed fastly, more
It is easy the advantages that carrying out image enhancement and display, has been widely used for various medical inspections and diagnosis at present.
In order to improve the resolution ratio of DR image, most of DR products can all receive in X-ray and be added one between plate and human body
Block antiscatter grid.The effect of the grid is while allowing pass through from the positive X-ray across human body, so that from people
The X-ray that volume scattering comes out is filtered out, so that the same sensitivity speck on plate be avoided to receive the X from human body different tissues as far as possible
Optical signal, to improve picture contrast and spatial resolution.When the spatial frequency of grid is not the spatial sampling frequencies of image
Integral multiple when, the noise of striated can be shown on image, is grid shadow 13 as shown in Figure 1, typically appears as periodicity
Noise.This phenomenon is referred to as Moire effect (Moir é effect)；Generated striped is also known as Morie fringe or moire.It rubs
The presence of your striped will affect normal human tissue and show, and then influence diagnosis of the doctor to illness.In order to avoid this mole
Following several methods usually can be used in striped, it may be assumed that 1) line density for adjusting grid, so that its frequency is plate sample frequency
The integral multiple of (i.e. the inverse at pixel interval)；2) using movement grid, when the movement velocity of grid is sufficiently large, so that grid pass through pixel
When the time of point is less than the time for exposure of the pixel, grid shadow will weaken or disappear；3) this is removed using the method for image procossing
Kind moire.Wherein, method 1) and method 2) higher to the structure and installation requirement of grid, and method 3) to the structure of grid
It requires to reduce with mounting means, but needs to remove the striped generated when imaging, and requirement cannot be lacked by Imaging Plate itself
Falling into (such as bad line, bad point) bring influences.
Method for using image procossing, most of directly pressed down using based on spatial domain convolutional filtering or frequency domain
System.This kind of methods are easy to be influenced by Imaging Plate bad line, generate the ring effect of diffusion type in filtering around bad line
It answers, the image after the low-pass filtering of bad line image referring to fig. 2 and Fig. 3.In order to reduce this ringing effect, usually there are two types of done
Method.First is that being imitated when designing filtering in the order for reducing filter as far as possible by sacrificing frequency characteristic to exchange lower ring for
It answers.Second is that minimizing the mutation of surrounding pixel point at bad line or bad point, that is, the interpolation algorithm used calculates bad line or bad
The probable value of point.Because using common interpolation method such as linear interpolation, cubic interpolation etc. to carry out interpolation at bad line, can only obtain
To smooth transition, the periodicity of grid shadow is destroyed, frequency domain filtering is caused to generate ringing effect and cannot reach and vibration is effectively reduced
The effect of bell.For the DR image with grid shadow, first method be usually unable to reach thoroughly filter out grid shadow while without
Ringing effect is generated at bad line, and second method then needs to consider while interpolation the influence of grid shadow, i.e., cannot only consider
Simple neighbor interpolation should also take into account the periodic regularity of grid shadow.However, there has been no research, provide can be effective at present
Remove the algorithm that grid shadow simultaneously again can minimize the generation of ring.
Summary of the invention
According to an aspect of the present invention, the method that a kind of pair of image carries out noise suppressed is provided, comprising:
Image acquisition step: obtaining image to be processed, and the image to be processed is the image interfered by periodic noise；
Period determines step: determining the period of periodic noise in the image to be processed；
Optimization Solution step: the period of the image data of image to be processed and the periodic noise is substituted into Signal separator
Model optimizes signals separating model, obtains the clean picture signal corresponding to the image to be processed.
According to another aspect of the present invention, the device that a kind of pair of image carries out noise suppressed is provided, comprising:
Image collection module, for obtaining image to be processed, the image to be processed is the figure interfered by periodic noise
Picture；
Period determines step: determining the period of periodic noise in the image to be processed；
Optimization Solution step: the period of the image data of image to be processed and the periodic noise is substituted into Signal separator
Model optimizes signals separating model, obtains the clean picture signal corresponding to the image to be processed.
The method for carrying out noise suppressed to image of the invention is realized by signals separating model, and the model is being used
When combine period of the periodic noise determined and handled so that isolating periodic noise signal, aperiodicity
It is contemplated that the cyclic effects of noise when noise signal and these three components of clean picture signal, so that in clean image
Not comprising ringing effect, and then improve picture quality.
Detailed description of the invention
Fig. 1 is the schematic diagram that grid shadow generates when antiscatter grid works, wherein 11 receive plate for X-ray, 12 be filter line
Grid, 13 be grid shadow, and short arrow 14 is direct rays, and long arrow 15 is scattered rays；
Fig. 2 and Fig. 3 is respectively the image schematic diagram after the bad line image schematic diagram that bad line ring generates and low-pass filtering；
Fig. 4 is the flow diagram of the DR image grid shadow suppressing method with grid of an embodiment of the present invention；
Fig. 5 is that the DR picture signal decomposition inputted in an embodiment of the present invention is the schematic diagram of three signal components；
Fig. 6 is the example schematic of DCT frequency domain and its filter coefficient in an embodiment of the present invention；
Fig. 7 is the schematic diagram of signal component p shown in Fig. 5；
Fig. 8 is the flow diagram of the grid shadow angle detecting algorithm in an embodiment of the present invention；
Fig. 9 a and 9b are the schematic diagram being segmented in an embodiment of the present invention to image to be processed.
Specific embodiment
The present invention proposes that a kind of pair of image carries out noise suppressing method, and particularly, which is to be interfered by periodic noise
Image.This method can be applied to the digital radiation image (abbreviation DR image) including but not limited to antiscatter grid
In.Below for carrying out the inhibition of grid shadow (i.e. periodic noise is grid shadow) to the DR image with grid, to pair of the invention
Image carries out noise suppressing method and is described in detail, it should be appreciated that the following method can be suitably used for various needs and inhibit periodically
The image of noise.
The method for carrying out noise suppressed to image that the embodiment of the present invention proposes is based on the signal under such a situation
Processing: addition interference of the clean signal x by periodic noise p, and in observation since defect sensor introduces
Sparse noise s (also referred to as aperiodicity noise) needs to export d from sensor at this time to carry out estimating for tri- components of s, p and x
Meter.In short, tri- unknown components of s, p and x are found out according to formula d=s+p+x respectively, by solving formula d=s+p+x
Can isolate periodic noise signal (i.e. p-component, be grid shadow caused by noise), aperiodicity noise signal (i.e. s component,
For the noise as caused by bad line or bad point) and clean picture signal (i.e. x-component, for eliminate grid shadow and bad line bad point etc. it
Picture signal afterwards), so as to obtain clean picture signal.
Below by way of specific embodiment combination attached drawing, invention is further described in detail.
Embodiment 1:
As shown in figure 4, the present embodiment proposes a kind of grid shadow suppressing method with grid DR image, pass through party's body of laws
Reveal the method proposed by the present invention to image (image especially interfered by periodic noise such as grid shadow) progress noise suppressed,
Its general process is: determining the grid shadow period first, then carries out the inhibition of grid shadow (bold portion in diagram).
Determination for the grid shadow period usually can obtain the cycle parameter there are two types of method, first is that passing through range estimation image
And obtain, second is that being calculated automatically using numerical method, a kind of specific implementation about the latter can refer to implementation hereinafter
Example 4.
After the grid shadow period estimated, grid shadow inhibition can be carried out.Signal processing side above-mentioned is based in the present embodiment
Formula carries out the inhibition of grid shadow, i.e., signal d (image i.e. to be processed) is separated into three components (see Fig. 5), including unknown clean figure
As signal x, periodic noise p and aperiodicity noise s, these three components meet x+s+p=d, construct Signal separator mould accordingly
Type (abbreviation SPX model), and grid shadow period substitution SPX model is optimized, so as to obtain clean picture signal
x.In embodiment, the acquisition of SPX model can be constructed by the following method and be obtained；Either call directly prior stored mould
Type, the stored model can be as follows building in advance and obtain.
Here constructed SPX model includes cost function and constraint condition, and wherein cost function is to meet constraint condition
And to the relevant function of at least one of clean signal component, periodic signal component and nonperiodic signal component, clean signal
Component is the measure function of the smoothness about clean picture signal, and periodic signal component is about periodic noise and its period
Measure function, nonperiodic signal component is measure function about aperiodicity noise.
In the present embodiment, the expression formula of clean signal component isThe expression formula of periodic signal component is
The expression formula of nonperiodic signal component is e3||diag(w)s||1；Constraint condition includes fixed constraint condition, or including fixing
Constraint condition and specified constraint condition, wherein fixed constraint condition is x+s+Mu=d, Mu=p, specifies constraint condition and relates to here
And expression formula in the meaning of each parameter will be described below.
The cost function expression formula of the present embodiment are as follows:
Subject to x+s+Mu=d, (2)
Wherein Mu is periodic noise, and M is transition matrix, and u is the variable for rebuilding periodic signal, e1、e2And e3Respectively
For the weight of the size for controlling clean picture signal, periodic noise and aperiodicity noise, Qc=diag (c) Q, Q are
The amplitude spectrum matrix of image to be processed, c are weight vectors, w be for reducing or improve aperiodicity noise value weight.
In other embodiments, the cost item in cost function can become bound term, such as: cost function can be expressed asAnd the cost function also meets fixed constraint condition and specified constraint condition, which is
And e3||diag(w)s||1< r2Or the specified constraint condition isAlternatively, cost function
It can be expressed asAnd meet fixed constraint condition and specified constraint condition, which isAnd
e3||diag(w)s||1< r2Or the specified constraint condition isAlternatively, the cost letter
Number can be expressed as e3||diag(w)s||1, and meet fixed constraint condition and specified constraint condition, which isOr the specified constraint condition isAlternatively, the cost function can
It is expressed asAnd meet fixed constraint condition and specified constraint condition, which is e3||diag
(w)s||1< r2；Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint
Condition, the specified constraint condition areAlternatively, the cost function isAnd meets and fix
Constraint condition and specified constraint condition, the specified constraint condition areWherein, r1、r2、r3、r4、r5And r6It is default
It is worth such as empirical value or test value.
In the present embodiment, the measure function to s, p and x is using L1Or L2Norm, such as to s (i.e. aperiodicity noises
) use L1Norm uses L to x (i.e. clean picture signal items) and p (i.e. periodic noise item)2Norm；Certainly, at it
In its embodiment, L can be substituted using other obvious approximate or equivalent measures1Norm and/or L2Norm.For example, can
To use L to p (i.e. periodic noise items)1Norm, to x (i.e. clean picture signal items) and s (i.e. aperiodicity noises
) use L2Norm is also possible to s (i.e. aperiodicity noise items) using L1Norm, to x (i.e. clean picture signals
) and p (i.e. periodic noise item) using L2Norm, etc., and so on.
In the present embodiment, if Wherein ncFor matrix Q line number and not equal to zero, α and β point
Wei not be for controlling the weight of the size of periodic noise and aperiodicity noise and clean picture signal, K is periodically to make an uproar
The number of segment of sound segmentation and setting, t is the period of each section of periodic noise, n=KtT0, T0For t in each section of periodic noise
Number.
By taking the cost function of the present embodiment as an example (i.e. formula (1)), the SPX model of building is optimized, is to solve for
Following mathematical optimization problem, it is possible thereby to realize that grid shadow inhibits:
Subject to x+s+Mu=d (4)
I.e. SPX model is to minimize cost function as optimization aim, under the premise of meeting constraint condition, in conjunction with periodicity
The period of noise obtains clean picture signal, periodic noise and aperiodicity noise as parameter, by solving cost function.
Cost function is at this timeThen the mathematical optimization problem is to ask to make cost function most
The value of x, u, s of hour, and its constraint condition is x+s+Mu=d.Wherein L is used to s1Norm, L1The calculation method of norm is
The sum of the absolute value of amount of orientation each element uses L to x and u2Norm, L2The calculation method of norm is amount of orientation each element
Quadratic sum and then extraction of square root.It has been observed that in formula (3) and (4), d ∈ RnFor input signal (the pending noise inputted
Inhibit the DR image of processing), s is aperiodicity noise, and p=Mu is periodic signal, and x is clean signal.QcFor Weight
DCT (discrete cosine transform) matrix, it is therefore an objective to filter, i.e., curb the component of some frequency so that x minimums, calculate
Method is shown in formula (5).
Qc=diag (c) Q (5)
Wherein Q is that (it can be by figure for the DCT amplitude spectrum of input signal (i.e. the DR image of pending noise suppressed processing)
As seeking frequency spectrum after carrying out dct transform, and then acquire amplitude spectrum), c ∈ RnFor weight vectors (or filter coefficient vector).C's
Shape is set as the form of one or more wave crests, respectively corresponds the wave crest as caused by periodic noise, such as Fig. 6 on DCT frequency spectrum
It is shown.Filter coefficient is bigger, then bigger to the inhibition of periodic noise；When filter coefficient is 0, corresponding frequency spectrum will not
Decaying.ncIt is the equal of matrix Q for the number of the filter coefficient not equal to zerocLine number.Q in the specific implementationcFor cum rights
The DCT matrix of weight, Q in other specific implementationscIt can also be DFT (discrete Fourier transform) matrix or FFT of such as Weight
(Fast Fourier Transform (FFT)) matrix etc..
For the variable u for rebuilding periodic signal, each signal can be decomposed into K sections, and every section with a fundamental wave table
Show, the length of the fundamental wave is t, repeats T0Secondary, t is the period of grid shadow, can be obtained by empirical value, or by subsequent right
The algorithm for estimating in grid shadow period is calculated and is obtained.The purpose of this segmentation is to increase robustness.As shown in fig. 7, with the period shown in Fig. 5
For signal p, periodic signal p can be analyzed to K sections, and every section is indicated with a fundamental wave, and the length of the fundamental wave is t, repeat T0Secondary (figure
Show middle T0=3).U can be expressed as follows:
Wherein ui∈Rt
M is signal transition matrix, so thatWherein n, t, K, T0
There is following relationship: n=KtT0.When signal length is not tT0Integral multiple when, can to Mu carry out truncation.
w∈RnIt is input weight, by increasing or decreasing wiValue can reduce or improve noise item s in position i (i.e. si)
Value.α, β control the size relation of periodic signal p and noise item s and signal x respectively.Periodic signal p phase can be reduced by increasing α
For the amplitude of s and x, simultaneously because the relationship of equality constraint, i.e. formula (4), the corresponding depreciation of the amplitude of s and/or x.It is also same to β
Reason.Parameter w, α, β described herein etc. in actual use can based on experience value or test value obtains.
Above-mentioned formula (3) and (4) are related to L1Norm and L2The mathematical optimization of norm, can by existing related mathematical knowledge into
Row solves, and three components p, s and the x isolated can be obtained, x is the clean picture signal after separating, that is, eliminates grid
Picture signal after shadow and bad line bad point.
To sum up, the method for carrying out noise suppressed to the image interfered by periodic noise of the present embodiment is based on SPX mould
Type is realized, periodic noise signal, aperiodicity noise signal and clean picture signal can be isolated by the model, by the party
Method be applied to using the DR image with grid processing in, can in the grid shadow being effectively removed in image, and due to
The period that noise is combined when using SPX model is handled, so that isolating periodic noise signal, aperiodicity noise
When signal and clean picture signal, it is contemplated that the cyclic effects of noise, make not including ringing effect in clean image,
And then improve picture quality.
Based on the above method, the present embodiment additionally provides the device that a kind of pair of image carries out noise suppressed comprising:
Image collection module, for obtaining image to be processed, which is the image interfered by periodic noise；
Period determination module, for determining the period of periodic noise in image to be processed；
Optimization Solution module, for substituting into the period of the image data of image to be processed and the periodic noise determined
Signals separating model optimizes signals separating model, obtains believing corresponding to the clean image of the image to be processed
Number.
The specific implementation of above each module can refer in above method embodiment to be described accordingly, is not repeated here herein.
Embodiment 2:
The present embodiment is still illustrated for the carry out grid shadow inhibition with grid DR image, and assumes to deposit at bad line
In biggish noise, and other pixel noise fields are smaller or are that 0 (i.e. noise is big sparse noise large and sparse
Noise), so the present embodiment is before carrying out the inhibition of grid shadow, considered first to progress interpolation at aperiodicity noise to restore
Pixel value at this, to reduce the influence that aperiodicity noise inhibits grid shadow.As shown in figure 4, including the dotted line in diagram
The general process of part and bold portion, the grid shadow suppressing method with grid DR image that the present embodiment proposes is: obtaining first
Take grid shadow direction (the i.e. direction of periodic noise, it can be understood as be visual point of periodic noise presentation in the picture
Cloth direction), it is determined the need for carrying out interpolation according to grid shadow direction situation, estimates the period of grid shadow, then so as in grid shadow
The influence of this periodic component is considered when inhibition, and grid shadow is finally removed according to SPX model.
The method that image procossing can be used in the acquisition in grid shadow direction is detected to obtain automatically；It is of course also possible to pass through people
For mode obtain, such as by be manually entered, the available directional information of hardware switch.These artificial modes can join
It examines in such as human-computer interaction and is realized in such a way that interface inputs, it is not described here in detail.As for direction (the i.e. aperiodicity of bad line
The direction of noise, it can be understood as be the non-periodic noise such as visual distribution arrangement of bad line presentation in the picture), it can
To obtain by eye-observation, it is also possible to be calculated by way of calibration, specific scaling method can refer to existing
Scaling method, be not detailed herein.
It present embodiments provides a kind of method using image procossing and carries out grid shadow angle detecting automatically, so that will input
Image is divided into three types, i.e., there is no the image of grid shadow, first direction grid shadow, (i.e. the direction in the direction of grid shadow and bad line is flat
Row, referred to herein as lateral grid shadow) (i.e. the direction of grid shadow and the direction of bad line is parallel, this paper for image and second direction grid shadow
In be referred to as vertical grid shadow) image.In the present embodiment, the basic procedure of grid shadow angle detecting algorithm is as shown in figure 8, right first
The DR image zooming-out wavelet character of input, then using preparatory trained first support vector machines (Support Vector
Machine, SVM) classifier (classifier 1 in diagram) carries out having grid and the classification without grid；For the DR image for there are grid, will mention
The feature taken is again inputted into second SVM classifier (classifier 2 in diagram), judges that grid shadow is laterally or vertical.
The method of feature extraction involved in an embodiment of the present invention introduced below and SVM classifier design.
The purpose of feature extraction is that the feature vector for being conducive to the detection of grid shadow and the classification of grid shadow direction is extracted from image,
The present embodiment mainly extracts wavelet character and classifies, and extraction algorithm includes the following steps S11-S13:
Step S11 calculates two-stage two-dimensional wavelet transformation for input picture, Haar small echo or db small echo can be selected, obtain
To decomposition cV1, cV2 of decomposition cH1, cH2 and vertical direction of horizontal direction；
Step S12 decomposes obtained each of four decomposition, calculates horizontal and vertical ladder in such a way
Degree, that is, for each location of pixels (i, j), calculate the gradient of horizontal direction and vertical direction, count all pixels position Zhong Shui
Flat direction gradient is greater than the number of pixels N1 of the certain thresholding ε of vertical direction gradient, counts vertical direction ladder in all pixels position
Degree is greater than the number of pixels N2 of the certain thresholding ε of horizontal direction gradient, and N1 and N2 is normalized and is returned with image pixel number
It returns, two characteristic values as this decomposition；
Step S13 returns to four and decomposes feature vector of the eight obtained characteristic values as input picture.
SVM classifier is designed, the classifier of the present embodiment is obtained according to training sample training, and classifier parameters are main
Including supporting vector, the parameter of kernel function and biasing (bias).The formula of classification are as follows:
Wherein y is feature vector after the transformation of input picture, { xi}I=1 ... mFor supporting vector, aiFor weighting coefficient, b is
Biasing, g (yi, y) and it is kernel function.Each value of feature vector needs to input again after Pan and Zoom, i.e.,
WhereinFor the feature vector directly extracted from image, z and l are respectively zooming and panning parameter vector.
Classifier 1 uses rbf kernel function, i.e.,
Wherein σ is rbf parameter.Classifier 2 uses Polynomial kernel function, i.e.,
The parameter of classifier 1 includes supporting vector, weighting coefficient, biasing, rbf parameter.The parameter of classifier 2 includes supporting
Vector, weighting coefficient and biasing.These parameters are obtained in training.
Trained basic procedure can be described generally as: get out three parts sample, the respectively image without grid shadow first
The image pattern of sample, the image pattern of lateral grid shadow and vertical grid shadow；Then classifier 1 is trained, i.e., there will be grid shadow (including cross
To grid shadow and vertical grid shadow) image pattern as positive sample, marked as+1, the image pattern of no grid shadow is as negative sample, mark
Number it is -1, extracts the feature of positive negative sample respectively by preceding feature extraction algorithm, input classifier training algorithm is trained, and is obtained
To the parameter of classifier 1；Classifier 2 is finally trained, i.e., using the image pattern of lateral grid shadow as positive sample, marked as+1, is erected
To grid shadow image pattern as negative sample, marked as -1, extract the feature of positive negative sample respectively by preceding feature extraction algorithm,
Input classifier training algorithm is trained, and obtains the parameter of classifier 2.
When online classification, spy is extracted by preceding feature extraction algorithm for the image (i.e. aforementioned image to be processed) newly inputted
Sign, according to Fig. 8 process carry out whether there is or not grid shadow detection and grid shadow angle detecting.The discrimination formula of classifier is shown in above-mentioned formula (6).
It is judged as there is grid shadow when the output of classifier 1 calculated according to formula (6) is greater than 0, is judged as no grid shadow when less than zero.When point
The output of class device 2 calculated according to formula (6) is judged as lateral grid shadow when being greater than 0, is judged as vertical grid shadow when less than 0.It needs
It is noted that the label of positive sample or negative sample can be interchanged when training, correspondingly the obtained label of online classification is also answered
The exchange.
The wavelet character and combination SVM used in the present embodiment is classified, and there is shown specific wavelet characters to be
It is obtained by two-stage two-dimensional wavelet transformation and the kernel function of specific SVM classifier, it will be understood that in other embodiments
In, the statistical of Gabor wavelet feature, feature can also be such as used by changing the type of the wavelet transformation of the present embodiment
Extract feature, can also support vector machines to the present embodiment carry out the core, the core that including but not limited to change support vector machines
Function etc..In addition, in other embodiments, the detection algorithm in grid shadow direction can also be using known characteristics of image if not
Bending moment etc., can also be using known classifier such as neural network, fisher classifier etc., as long as whether there is or not grid shadows suitable for obtaining
And the feature in grid shadow direction and classification.
After obtaining grid shadow direction by the above method, grid shadow direction is judged.If vertical grid shadow (i.e. grid shadow
Direction is vertical with the direction of bad line), then conventional difference processing such as linear interpolation is directly carried out, carries out grid shadow period meter after interpolation
It calculates and subsequent grid shadow inhibition processing, it will be understood that be equivalent to that there is no bad lines in the DR image after interpolation, thus subsequent
Relevant weight can be set in processing accordingly；If lateral grid shadow (i.e. the direction of grid shadow and the direction of bad line is parallel), then not
It needs to interpolation processing is carried out at bad line, directly the progress grid shadow period calculates and subsequent grid shadow inhibition is handled.Herein further
Illustrate, when the aperiodicity noise detected is bad point, regardless of grid shadow direction, thinks the direction of grid shadow at this time
(as lateral grid shadow) parallel with bad point, directly progress grid shadow period calculate and subsequent grid shadow inhibition processing.
The correlation that the grid shadow cycle determination method and grid shadow suppressing method of the present embodiment can refer in previous embodiment 1 is retouched
It states.The calculating in grid shadow direction is one of the influence that grid shadow inhibits, due to judging that grid shadow direction is vertical with bad line direction
When, the pixel value at bad line is restored by interpolation method, so, when carrying out the inhibition of grid shadow, the value of input weight w above-mentioned
It can not have to be set as very big.
To sum up, the present embodiment not only has the advantages that embodiment 1, but also detects grid shadow direction first in processing, according to
Grid shadow direction determines whether to carry out bad line interpolation to restore the pixel value at bad line, has minimized bad line and has pressed down to subsequent gate shadow
The influence of system.
Method based on the present embodiment, the present embodiment additionally provide the device that a kind of pair of image carries out noise suppressed, packet
It includes:
Image collection module, for obtaining image to be processed, which is the image interfered by periodic noise；
Angle detecting module, for determining the direction of the periodic noise；
Period determination module, for determining the period of periodic noise in image to be processed；
Optimization Solution module, for substituting into the period of the image data of image to be processed and the periodic noise determined
Signals separating model optimizes signals separating model, obtains believing corresponding to the clean image of the image to be processed
Number.
The specific implementation of above each module can refer in above method embodiment to be described accordingly, is not repeated here herein.
Embodiment 3:
In embodiment 1 or 2, it is assumed that in the cost item about x of cost function, QcFor DCT matrix, i.e. Qc∈Rn×n, with
The increase of the every a line length n of image, the computation complexity of the cost function based on SPX model it is in square increasing, i.e., progressive multiple
Miscellaneous degree is O (n2).It is appreciated that QcFor FFT matrix or DFT matrix etc., there is also similar complexity issues.The present embodiment is
It is improved for possible higher algorithm complexity existing for embodiment 1 or 2, it is therefore an objective to reduce the algorithm based on SPX model
Complexity.
In the present embodiment, to reduce complexity, handled by the way of segmentation, it is assumed that be divided into nsSection, then it is progressive multiple
Miscellaneous degree becomesTherefore, work as nsWhen closer to n, the complexity of algorithm is close to O (n), i.e., linear complicated
Degree.It should be understood that the segmentation said in the present embodiment is different from the segmentation of variable u in embodiment 1, the segmentation in the present embodiment refers to pair
A line image data is segmented, and is then calculated again using the algorithm in embodiment 1, by 1 knot of the present embodiment and embodiment
It closes, it is obvious that the targeted data processing of the segmentation of variable u is to have been segmented into obtain based on the present embodiment in embodiment 1 at this time
A certain section of image data.
Generally, it if using segment processing, needs to solve to be segmented side caused by the side effect due to frequency domain filtering
The discontinuous problem on boundary.In this regard, the present embodiment handles border issue in this way: a line image data is divided into nsSection, each section
Front and back reserves sub-fraction as overlapping region, and last image data is from a cycle of each section of discarding and last
The combination of a cycle.As illustrated in figures 9 a and 9b, the present invention can be shown to the segment processing of image data.In fig. 9 a,
A line image data 90 is segmented, two sections of adjacent image data two periods of overlapping of front and back, it is thick black in lap such as figure
Line part and most shown in heavy black part.It is each in addition to first segment image data and final stage image data in fragmentation procedure
Section image data abandons its a cycle and the last one period.By taking second segment and third section image data as an example, second
Section abandons heavy black section of the lap marked as 201 and the heavy black section marked as 204, and third section abandons lap label
For 301 and the heavy black section marked as 304.The result of above-mentioned segment processing is shown in Fig. 9 b, although second segment and third section are lost respectively
First section and end overlapping are abandoned, but since second segment has been supplied the period 301 abandoned, third section has supplied the period 204 abandoned,
In two sections of junction image datas and discontinuous phenomenon is not present.Further as shown in figure 9b, except first segment image data
Several block numbers evidences are originated with several block numbers in end of final stage image data outside, several pieces of the starting of each section of image data
Data (label is heavy black part in diagram) and several block numbers evidences in end of the last period image data (mark in diagram and are
Heavy black part) overlapping, several block numbers in the end of each section of image data are according to (label is heavy black part in diagram) with after
Several block numbers of the starting of one section of image data are according to (label is heavy black part in diagram) overlapping.
As a whole, the general process of the noise suppressing method to the image interfered by periodic noise of the present embodiment
It is: first to the image to be processed of input interfered by periodic noise, determines it by previous embodiment 1 (or embodiment 2)
The image to be processed is segmented by the period (or direction and period) of periodic noise by the segmentation method of the present embodiment
Processing, obtains multistage image data to be processed, then for each section of image data to be processed using the grid shadow suppression in embodiment 1
Method processed is handled, the result after finally obtaining the noise suppressed of image to be processed.It has been observed that due to the side for using segmentation
Formula reduces the complexity of algorithm.
Embodiment 4:
A kind of grid shadow suppressing method with grid DR image that the present embodiment proposes can be based on foregoing embodiments and realize,
But be directed to really the fixed grid shadow period using a kind of algorithm based on SPX model realization.
The size in grid shadow period is based on SPX model above-mentioned in embodiment and its mathematical optimization problem is realized, can be by such as
Lower algorithm obtains:
For the image to be processed of input, chooses one of area-of-interest and carry out intermittent scanning, find optimal week
Phase, candidate periodic are empirical value, such as can be set to the range of 5-10, or be adjusted according to the actual situation.Region of interest
The region that domain preferably answers selection cycle obvious is avoided due to underdosage or the superfluous region for causing periodic signal too weak,
Or by the way of randomly selecting.Since the image comprising grid shadow is after Fourier (DFT) transforms to frequency domain, grid shadow is in frequency
It is distributed in multiple frequencies (high frequency, intermediate frequency, low frequency) in domain to be overlapped with the component frequency after image information conversion, causes on frequency spectrum
It can be seen that several Frequency point Energy distributions are high compared with around it, so the higher place of this Frequency and Amplitude is accordingly locating for grid shadow
Frequency field；Based on this, grid shadow zone domain can choose out, i.e. period obvious region, which is area-of-interest.
The balancing method of optimal period are as follows: the periodic signal finally calculated is stronger, then the period is closer to the true period；
That is the L of calculating cycle signal2Norm | | p | |2。L2The calculation method of norm are as follows: the quadratic sum of amount of orientation each element opens radical sign again.
Therefore, taking the strongest model of the periodic signal corresponding period for decompositing and is the period of grid shadow.With aforementioned in the present embodiment
It for the cost function of formula (3), slightly deforms, performance mathematically is:
It calculates first
Subject to x+s+Mu=d (11)
What wherein argmin was indicated is variate-value when being minimized objective function, is embodied in that this refers to asking makes generation
U value when valence function is minimized；
Then it is calculated according to the u value being calculated
In the formula and expression formula referred here to, what d was indicated is the image data of area-of-interest；What t was indicated is pre-
The candidate periodic first set, by preset candidate periodic within the scope of some substitute into the formula herein being related to and expression formula into
Row calculates, and solving that the corresponding candidate periodic of the strongest model of periodic noise signal come is optimal period, that is, needs
The period for the periodic noise sought；D is difference matrix, and form is as follows:
The effect of difference matrix is the difference for seeking signal, i.e. Dx=[x1-x2, x2-x3..., xn-1-xn]T, wherein | |
Dx||1Referred to as total variance (total-variation) is for measuring estimating for picture signal smoothness；N is region of interest
The length of the image data in domain；The meaning of other parameters is not repeated here herein referring to previous embodiment.It can when due to cycle estimator
To avoid at bad line, weight vectors w can be set to complete 1.It is appreciated that slightly become by taking the majorized function of formula (3) as an example here
Shape carries out the explanation of period calculating, and the majorized function can be foregoing cost item in other embodiments can become about
Shu Xiang, the coefficient before each cost item can accordingly modify and L2Norm is also readily modified as other equivalent measures etc., as long as its
It is suitable for calculating the function in preferable period.
In order to increase robustness, multirow can be chosen and calculated, look for the cumulative of the quadratic sum of the norm in all row periods
It is worth grid shadow cycle estimator of the maximum corresponding period as present image.
According to the present embodiment, an embodiment of the present invention additionally provides the period of periodic noise in a kind of determining image
Method, comprising:
Image acquisition step: obtaining image to be processed, and the image to be processed is the image interfered by periodic noise, from
The image data of area-of-interest is obtained in the image to be processed；
Optimization Solution step: the image data of area-of-interest and preset candidate periodic are substituted into Signal separator mould
Type optimizes signals separating model, obtains the clean image letter of the image data corresponding to the area-of-interest
Number, periodic noise and aperiodicity noise；
Period determines step: the corresponding candidate periodic of the strongest model of periodic noise signal solved is determined as week
The period of phase property noise.
The specific implementation of above each step can refer to the associated description in the period of aforementioned determining periodic noise, herein not
It repeats.
Based on the embodiment of the method in the period of periodic noise in the determination image, another embodiment additionally provides one
Kind determines the device in the period of periodic noise in image, comprising:
Image collection module, for obtaining image to be processed, the image to be processed is the figure interfered by periodic noise
Picture obtains the image data of area-of-interest from the image to be processed；
Optimization Solution module, for the image data of area-of-interest and preset candidate periodic to be substituted into signal point
From model, signals separating model is optimized, obtains the clean figure of the image data corresponding to the area-of-interest
As signal, periodic noise and aperiodicity noise；
Period determination module, the corresponding candidate periodic of the strongest model of periodic noise signal for will solve determine
For the period of periodic noise.
The specific implementation of above each module can refer to the associated description in the period of aforementioned determining periodic noise, herein not
It repeats.
Based on above embodiments, the noise suppressing method or dress proposed by the present invention to the image interfered by periodic noise
In setting, periodic noise can be isolated by passing through building SPX model and optimizing, moreover it is possible to isolate sparse noise
(i.e. aperiodicity noise), so as to obtain clean picture noise；In addition, by segment processing mode in a kind of embodiment,
It can significantly reduce the algorithm complexity based on SPX model, and the boundary effect of segmentation can be effectively eliminated.According to the embodiment of the present invention
To image carry out noise suppressed method and device thereof, can be that present X is penetrated by hardware, software, firmware or a combination thereof
In line imaging system, so that the method that x-ray imaging system can use the noise suppressed according to the embodiment of the present invention, or
Person includes the device according to the noise suppressed of the embodiment of the present invention.
It will be understood by those skilled in the art that all or part of the steps of various methods can pass through in above embodiment
Program instructs related hardware to complete, which can be stored in a computer readable storage medium, storage medium can wrap
It includes: read-only memory, random access memory, disk or CD etc..
The foregoing is a further detailed description of the present application in conjunction with specific implementation manners, and it cannot be said that this Shen
Specific implementation please is only limited to these instructions.For those of ordinary skill in the art to which this application belongs, it is not taking off
Under the premise of from the application design, a number of simple deductions or replacements can also be made.
Claims (14)
1. the method that a kind of pair of image carries out noise suppressed characterized by comprising
Image acquisition step: obtaining image to be processed, and the image to be processed is the image interfered by periodic noise；
Period determines step: determining the period of periodic noise in the image to be processed；
Optimization Solution step: the period of the image data of image to be processed and the periodic noise is substituted into Signal separator mould
Type optimizes signals separating model, obtains the clean picture signal corresponding to the image to be processed；
Wherein, the signals separating model includes cost function and constraint condition, and the cost function is to meet the constraint item
Part and to the relevant function of at least one of clean signal component, periodic signal component and nonperiodic signal component；
The clean signal component is the measure function of the smoothness about clean picture signal, and the periodic signal component is to close
In periodic noise and its measure function in period, the nonperiodic signal component is to estimate letter about aperiodicity noise
Number.
2. the method according to claim 1, wherein the constraint condition includes fixed constraint condition, Huo Zhesuo
Stating constraint condition includes fixed constraint condition and specified constraint condition,
The fixed constraint condition are as follows: x+s+Mu=d,
The expression formula of the clean signal component includes:
The expression formula of the periodic signal component includes:
The expression formula of the nonperiodic signal component includes: e3||diag(w)s||1,
Wherein d indicates that image to be processed, x indicate clean picture signal, and s indicates that aperiodicity noise, u are indicated for rebuilding the period
Property noise variable, Mu indicate periodic noise, M is transition matrix, e1、e2And e3Respectively for control clean picture signal,
The weight of the size of periodic noise and aperiodicity noise, Qc=diag (c) Q, Q are the amplitude spectrum matrix of image to be processed, c
For weight vectors, w be for reducing or improve aperiodicity noise value weight；
The cost function isAnd meet fixed constraint condition；
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, the specified constraint
Condition isAnd e3||diag(w)s||1< r2Or the specified constraint condition is
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, the specified constraint item
Part isAnd e3||diag(w)s||1< r2Or the specified constraint condition is
Alternatively, the cost function is e3||diag(w)s||1, and meet fixed constraint condition and specified constraint condition, the finger
Determining constraint condition isAndOr the specified constraint condition is
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, the finger
Determining constraint condition is e3||diag(w)s||1< r2；
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition,
The specified constraint condition is
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, institute
Stating specified constraint condition is
Wherein, r1、r2、r3、r4、r5And r6For preset value.
3. according to the method described in claim 2, it is characterized in that,ncFor matrix Q's
Line number and the size for respectively being used to control periodic noise and aperiodicity noise and clean picture signal not equal to zero, α and β
Weight, K be by periodic noise be segmented and set number of segment, t be each section of periodic noise period, n=KtT0, T0For
The number of t in each section of periodic noise.
4. the method according to claim 1, wherein the step for determining the period of the periodic noise it
Before, the method also includes: determine the direction of the periodic noise, if it is determined that go out the periodic noise direction with it is non-
The direction of periodic noise is vertical, carries out interpolation processing to the image to be processed, continues to execute the period and determines step, if really
The direction for making the periodic noise is parallel with the direction of aperiodicity noise, and the directly execution period determines step；
Wherein the step for the direction of the determining periodic noise specifically:
Feature extraction is carried out to the image to be processed；
According to the feature extracted, classified using the first classifier that preparatory training obtains to the image to be processed, it will
The image recognition to be processed is the image with periodic noise or the image without periodic noise, and first classifier is
By the way that the image with periodic noise is included in positive sample training set, negative sample training will be included in without the image of periodic noise
Training get or first classifier be by the way that the image with periodic noise is included in negative sample training set, will not
Image with periodic noise is included in the training of positive sample training set and obtains；
According to the feature extracted, the second classifier obtained using preparatory training is to obtained band periodic noise of classifying
Image classify, by the image recognition with periodic noise be first direction noise image and second direction noise pattern
Picture, wherein the direction of periodic noise is parallel with the direction of aperiodicity noise in the first direction noise image, described
The direction of periodic noise is vertical with the direction of aperiodicity noise in two direction noise images, and second classifier is to pass through
First direction noise image is included in positive sample training set, that second direction noise image is included in negative sample training set is trained
It arrives or second classifier is by the way that first direction noise image is included in negative sample training set, by second direction noise
Image is included in the training of positive sample training set and obtains.
5. according to the method described in claim 4, it is characterized in that, the feature extracted includes wavelet character；Described
One classifier and second classifier are support vector machine classifier.
6. the method according to claim 1, wherein the method also includes dividing the image to be processed
Section processing, obtains multistage image data, for each section of image data to be processed, executes the Optimization Solution step；
Wherein, the segment processing includes: that image to be processed is divided into multistage, wherein if except the starting of first segment image data
Dry block number according to and final stage image data several block numbers in end according to outer, several block numbers of the starting of each section of image data according to
Several pieces of the end of the last period image data data overlap, several block numbers evidences in the end of each section of image data and latter section of image
Several pieces of data overlaps of the starting of data.
7. the method according to claim 1, wherein the period of the periodic noise is confirmed as preset value,
Or the period of the periodic noise includes: by following steps calculating determination
Obtain the image data of area-of-interest；
The image data of area-of-interest and preset candidate periodic are substituted into signals separating model, to signals separating model
Clean picture signal, periodic noise and aperiodicity noise optimization is carried out to solve；
The corresponding candidate periodic of the strongest model of periodic noise signal solved is determined as to the period of periodic noise.
8. the method according to the description of claim 7 is characterized in that
It is described that the formula that clean picture signal, periodic noise and aperiodicity noise optimization solve is carried out to signals separating model
Include:
And about
Beam condition is x+s+Mu=d；
The corresponding candidate periodic of the strongest model of periodic noise signal come that will solve is determined as periodic noise
The step for period corresponding expression formula are as follows:
Wherein, d indicates the image data of area-of-interest, and x indicates clean picture signal, and s indicates that aperiodicity noise, u indicate
For rebuilding the variable of periodic noise, Mu indicates periodic noise, and M is transition matrix, and D is difference matrix, w be for reducing
Or the weight of the value of aperiodicity noise is improved, n is the length of the image data of the area-of-interest, and α and β are respectively to be used for
Control the weight of the size of periodic noise and aperiodicity noise and clean picture signal, K be by periodic noise segmentation and
The number of segment of setting, t are the candidate periodic, n=KtT0, T0For the number of t in each section of periodic noise,For the period
The period of property noise.
9. the method as described in claim 1, which is characterized in that the image to be processed includes digital radiation image, the week
Phase property noise includes grid shadow.
10. the device that a kind of pair of image carries out noise suppressed characterized by comprising
Image collection module, for obtaining image to be processed, the image to be processed is the image interfered by periodic noise；
Period determination module, for determining the period of periodic noise in the image to be processed；
Optimization Solution module, for the period of the image data of image to be processed and the periodic noise to be substituted into Signal separator
Model optimizes signals separating model, obtains the clean picture signal corresponding to the image to be processed；
Wherein, the signals separating model includes cost function and constraint condition, and the cost function is to meet the constraint item
Part and to the relevant function of at least one of clean signal component, periodic signal component and nonperiodic signal component；
The clean signal component is the measure function of the smoothness about clean picture signal, and the periodic signal component is to close
In periodic noise and its measure function in period, the nonperiodic signal component is to estimate letter about aperiodicity noise
Number.
11. device according to claim 10, which is characterized in that the constraint condition includes fixed constraint condition, or
The constraint condition includes fixed constraint condition and specified constraint condition,
The fixed constraint condition are as follows: x+s+Mu=d,
The expression formula of the clean signal component includes:
The expression formula of the periodic signal component includes:
The expression formula of the nonperiodic signal component includes: e3||diag(w)s||1,
Wherein d indicates that image to be processed, x indicate clean picture signal, and s indicates that aperiodicity noise, u are indicated for rebuilding the period
Property noise variable, Mu indicate periodic noise, M is transition matrix, e1、e2And e3Respectively for control clean picture signal,
The weight of the size of periodic noise and aperiodicity noise, Qc=diag (c) Q, Q are the amplitude spectrum matrix of image to be processed, c
For weight vectors, w be for reducing or improve aperiodicity noise value weight；
The cost function isAnd meet fixed constraint condition；
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, the specified constraint
Condition isAnd e3||diag(w)s||1< r2Or the specified constraint condition is
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, the specified constraint item
Part isAnd e3||diag(w)s||1< r2Or the specified constraint condition is
Alternatively, the cost function is e3||diag(w)s||1, and meet fixed constraint condition and specified constraint condition, the finger
Determining constraint condition isAndOr the specified constraint condition is
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, the finger
Determining constraint condition is e3||diag(w)s||1< r2；
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition,
The specified constraint condition is
Alternatively, the cost function isAnd meet fixed constraint condition and specified constraint condition, institute
Stating specified constraint condition is
Wherein, r1、r2、r3、r4、r5And r6For preset value.
12. device according to claim 10, which is characterized in that further include: angle detecting module, for determining the week
The direction of phase property noise；
The wherein direction of the determining periodic noise specifically:
Feature extraction is carried out to the image to be processed；
According to the feature extracted, classified using the first classifier that preparatory training obtains to the image to be processed, it will
The image recognition to be processed is the image with periodic noise or the image without periodic noise, and first classifier is
By the way that the image with periodic noise is included in positive sample training set, negative sample training will be included in without the image of periodic noise
Training get or first classifier be by the way that the image with periodic noise is included in negative sample training set, will not
Image with periodic noise is included in the training of positive sample training set and obtains；
According to the feature extracted, the second classifier obtained using preparatory training is to obtained band periodic noise of classifying
Image classify, by the image recognition with periodic noise be first direction noise image and second direction noise pattern
Picture, wherein the direction of periodic noise is parallel with the direction of aperiodicity noise in the first direction noise image, described
The direction of periodic noise is vertical with the direction of aperiodicity noise in two direction noise images, and second classifier is to pass through
First direction noise image is included in positive sample training set, that second direction noise image is included in negative sample training set is trained
It arrives or second classifier is by the way that first direction noise image is included in negative sample training set, by second direction noise
Image is included in the training of positive sample training set and obtains.
13. device according to claim 10, which is characterized in that the period of the periodic noise is confirmed as presetting
It is worth or the calculating in the period of the periodic noise includes:
Obtain the image data of area-of-interest；
The image data of area-of-interest and preset candidate periodic are substituted into signals separating model, to signals separating model
Clean picture signal, periodic noise and aperiodicity noise optimization is carried out to solve；
The corresponding candidate periodic of the strongest model of periodic noise signal solved is determined as to the period of periodic noise.
14. device as claimed in claim 10, which is characterized in that the image to be processed includes digital radiation image, described
Periodic noise includes grid shadow.
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2014/073461 WO2015135208A1 (en) | 2014-03-14 | 2014-03-14 | Method and device for conducting noise suppression on image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106464778A CN106464778A (en) | 2017-02-22 |
| CN106464778B true CN106464778B (en) | 2019-11-05 |
Family
ID=54070835
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201480077057.XA Active CN106464778B (en) | 2014-03-14 | 2014-03-14 | The method and device of noise suppressed is carried out to image |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN106464778B (en) |
| WO (1) | WO2015135208A1 (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108111240B (en) * | 2017-12-19 | 2020-10-27 | 中国科学院上海天文台 | Data processing method and device for radio frequency interference |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102262775A (en) * | 2010-05-26 | 2011-11-30 | 深圳迈瑞生物医疗电子股份有限公司 | Image denoising method and device and image processing equipment |
| CN102279971A (en) * | 2011-04-06 | 2011-12-14 | 中国人民解放军理工大学气象学院 | Variational method-based method for reducing stripe noise of moderate-resolution imaging spectroradiometer |
| CN102750683A (en) * | 2012-06-18 | 2012-10-24 | 常州大学 | Filtering method for sea surface stripe noise and sea surface stripe cloud in moderate resolution imaging spectroradiometer (MODIS) remote sensing image |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100547438C (en) * | 2007-07-10 | 2009-10-07 | 廊坊开发区中油油田科技工贸有限责任公司 | A kind of oil-gas exploration method and system |
| US8346008B2 (en) * | 2009-06-09 | 2013-01-01 | Aptina Imaging Corporation | Systems and methods for noise reduction in high dynamic range imaging |
| EP2491478A4 (en) * | 2009-10-20 | 2014-07-23 | Cypress Semiconductor Corp | Method and apparatus for reducing coupled noise influence in touch screen controllers. |
| CN103020913B (en) * | 2012-12-18 | 2015-06-24 | 武汉大学 | Remote-sensing image stripe noise removing method based on segmented correction |
-
2014
- 2014-03-14 WO PCT/CN2014/073461 patent/WO2015135208A1/en active Application Filing
- 2014-03-14 CN CN201480077057.XA patent/CN106464778B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102262775A (en) * | 2010-05-26 | 2011-11-30 | 深圳迈瑞生物医疗电子股份有限公司 | Image denoising method and device and image processing equipment |
| CN102279971A (en) * | 2011-04-06 | 2011-12-14 | 中国人民解放军理工大学气象学院 | Variational method-based method for reducing stripe noise of moderate-resolution imaging spectroradiometer |
| CN102750683A (en) * | 2012-06-18 | 2012-10-24 | 常州大学 | Filtering method for sea surface stripe noise and sea surface stripe cloud in moderate resolution imaging spectroradiometer (MODIS) remote sensing image |
Non-Patent Citations (1)
| Title |
|---|
| 基于灰度归一化的 HJ-1A星HSI图像条带噪声去除方法;兰穹穹;《国土资源遥感》;20120615(第2期);第45-49页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2015135208A1 (en) | 2015-09-17 |
| CN106464778A (en) | 2017-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104915930B (en) | The method and device of grey level compensation and noise suppressed is carried out to image | |
| Wang et al. | A new image denoising method based on Gaussian filter | |
| CN109859126A (en) | A kind of vedio noise reduction method, apparatus, electronic equipment and storage medium | |
| EP2819091B1 (en) | Method and apparatus for processing a gray image | |
| Li et al. | An adaptive method of speckle reduction and feature enhancement for SAR images based on curvelet transform and particle swarm optimization | |
| CN108399611B (en) | Multi-focus image fusion method based on gradient regularization | |
| CN107464226B (en) | A kind of image de-noising method based on improvement two-dimensional empirical mode decomposition algorithm | |
| Pattichis et al. | Analyzing image structure by multidimensional frequency modulation | |
| CN108573499A (en) | A kind of visual target tracking method based on dimension self-adaption and occlusion detection | |
| CN109003233A (en) | A kind of image de-noising method based on adaptive weighting Total Variation | |
| Moraru et al. | Dempster-shafer fusion for effective retinal vessels’ diameter measurement | |
| CN111768350A (en) | Infrared image enhancement method and system | |
| Zhou et al. | Nonlocal means filtering based speckle removal utilizing the maximum a posteriori estimation and the total variation image prior | |
| JP3406577B2 (en) | Object recognition method | |
| Zhang et al. | An improved BM3D algorithm based on anisotropic diffusion equation | |
| Nakanishi et al. | Anomaly detection by autoencoder based on weighted frequency domain loss | |
| CN106464778B (en) | The method and device of noise suppressed is carried out to image | |
| Gyger et al. | Three-dimensional speckle reduction in optical coherence tomography through structural guided filtering | |
| Liu et al. | Phases measure of image sharpness based on quaternion wavelet | |
| Wang et al. | Region-based adaptive anisotropic diffusion for image enhancement and denoising | |
| CN104867120B (en) | The non local method for reducing speckle of SAR image based on than Distribution value | |
| CN109658357A (en) | A kind of denoising method towards remote sensing satellite image | |
| Iskarous | Detecting the edge of the tongue: A tutorial | |
| Shi et al. | A novel approach of edge detection based on gray weighted absolute correlation degree and Prewitt operator | |
| Sun et al. | Small infrared target detection using frequency-spatial cues in a single image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210324 Address after: 518000 201 Genyu Road, Tianliao community, Yutang street, Guangming District, Shenzhen City, Guangdong Province Patentee after: Shenzhen Mindray Software Technology Co.,Ltd. Address before: 518057 MINDRAY building, twelve South tech Road, Nanshan District hi tech Industrial Park, Shenzhen, Guangdong Patentee before: Shenzhen Mindray Bio-Medical Electronics Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| CP02 | Change in the address of a patent holder |
Address after: 518000 514a, No. 1301-74, sightseeing Road, Xinlan community, Guanlan street, Longhua District, Shenzhen, Guangdong Province Patentee after: Shenzhen Mindray Software Technology Co.,Ltd. Address before: 518000 201 Genyu Road, Tianliao community, Yutang street, Guangming District, Shenzhen City, Guangdong Province Patentee before: Shenzhen Mindray Software Technology Co.,Ltd. |
|
| CP02 | Change in the address of a patent holder |