CN109919870A - A kind of SAR image speckle suppression method based on BM3D - Google Patents
A kind of SAR image speckle suppression method based on BM3D Download PDFInfo
- Publication number
- CN109919870A CN109919870A CN201910162471.5A CN201910162471A CN109919870A CN 109919870 A CN109919870 A CN 109919870A CN 201910162471 A CN201910162471 A CN 201910162471A CN 109919870 A CN109919870 A CN 109919870A
- Authority
- CN
- China
- Prior art keywords
- image
- sar image
- log
- domain
- sar
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention proposes a kind of SAR image Speckle Reduction Algorithm based on BM3D, the effective of coherent spot is inhibited and grain details are effectively retained the defect taken into account, implementation step to solve existing in the prior art can not achieve are as follows: L view SAR image progress natural logrithm converted to obtain log-domain SAR image I1, to I1It is pre-processed to obtain the log-domain SAR image I that coherent speckle noise mean value is zero3, calculate I3The standard deviation sigma of middle coherent speckle noise, using BM3D algorithm to I3It carries out Speckle reduction and obtains the log-domain SAR image I after Speckle reduction4, to I4It carries out inverse transformation and obtains I0Suppression spot image Iresult.The present invention effectively inhibit image coherent speckle noise while can be good at retain image grain details part, thus be effectively guaranteed SAR image it is subsequent classified, the accuracy of Target detection and identification.
Description
Technical field
The invention belongs to technical field of image processing, are related to a kind of SAR image speckle suppression method, and in particular to a kind of
SAR image speckle suppression method based on BM3D can be used for classifying to SAR image, Target detection and identification.
Background technique
Synthetic aperture radar (Synthetic Aperture Radar, SAR) is a kind of work being concerned in microwave band
Imaging radar, it becomes current remote sensing observations with the data retrieval capabilities of its high-resolution, round-the-clock, round-the-clock and large area
Important means, resource, environment, archaeology and in terms of be widely used.But special image-forming principle is given
SAR is brought while advantage but also the SAR image obtained endures the puzzlement of coherent speckle noise to the fullest extent, i.e., in the SAR image after imaging
Each point pixel value is not the true backscattering coefficient of target, but random violent fluctuation in its vicinity, and is different from general
The coherent speckle noise of the additive Gaussian noise of logical image institute band, SAR image is multiplicative noise and fluctuation range is larger, is seriously affected
The subsequent applications of SAR image, such as the practical application areas such as feature detection and target identification.Therefore, the coherent spot of SAR image
Inhibit to be always SAR image interpretation and the indispensable image processing module of application field.
Predominantly two for evaluating the index of SAR image speckle suppression method, one is equivalent number (Equivalent
Numbers of Looks, ENL), for evaluating SAR image speckle suppression method to the degree of Speckle reduction, ENL value is got over
Height indicates that suppression spot ability is stronger；The other is the ratio images of the SAR image after SAR original image and Speckle reduction, are used to
SAR image speckle suppression method is evaluated to the reserving degree of image texture details.The theory base according to used by distinct methods
SAR image speckle suppression method is divided into four classes by plinth: the airspace filter technology after multiple look processing, imaging, transformation before imaging
Domain filtering technique and non-local mean filtering technology.
Multiple look processing is the speckle suppression method to grow up at first, it is to carry out in SAR orientation to radar aperture
Segmentation, each aperture carries out imaging respectively, then will take average place after obtained multiple image non-coherent additions depending on number
Reason obtains the SAR image of Speckle reduction.The although simple, convenient realization of multiple look processing technology, but acquisition can be reduced
SAR image resolution ratio and to press down spot effect poor.
The representative that airspace filter technology most starts is traditional non-adaptive such as median filtering, mean filter, K Neighborhood Filtering
Filter, this kind of filter thought is simple, convenient for operation but effect it is poor.Develop later and has locally been united based on coherent speckle noise
Count information Lee filtering, Kuan filtering, Frost filtering etc. sef-adapting filters, then according to image pixel homogeneous region also
It is that the above-mentioned filter of texture region improvement has obtained enhanced Lee filtering, enhancing Frost filtering.In addition there are assume known SAR figure
As ground scatter coefficient obey Gamma distribution or other more accurate distributed models described in needing more parameters,
The filtering algorithm that true ground scene is estimated using maximum a posteriori probability (Maximum a Posteriori, MAP), most
Common is exactly Gamma MAP filtering.The advantages of this kind of algorithm is equally that thought is simple, facilitates realization, but generally existing relevant
Spot inhibits to be affected by local window parameter setting with grain details reservation and the problems such as noise suppressed is insufficient often occur.
The basic thought of transform domain filtering is the description target image for the multiresolution that wavelet transformation can be local, to make
The noise and signal obtained in image rough can be separated in wavelet field, and reach SAR image by a series of subsequent processing
The effect of Speckle reduction.The method of wavelet transformation has preferable reservation to the texture information of image, therefore by using not
Choosing suitable threshold value with strategy can achieve the grain details of reservation image on the basis of image Speckle reduction.Such as it is soft
Threshold value (Soft Threshoding, ST) filtering, by being carried out at adaptive threshold value to the coefficient after SAR image wavelet transformation
Reason achievees the purpose that Speckle reduction.But since wavelet transformation is to noise and signal can not be totally separated from so that this method
Suppression spot effect be not very sufficiently.
The thought of non-local mean (Non-local Means, NLM) filtering comes across natural image process field first,
Its basic thought is the autocorrelation using image, takes adding between them according to the degree of similarity of pixel surrounding pixel block
Image of the weight average to be denoised.NLM thought does not utilize the spatial relationship between image slices vegetarian refreshments, but utilizes them
Between autocorrelation, therefore hardly introduce deceptive information.The measurement of pixel autocorrelation is not only pixel simultaneously
Between point, but the similitude of pixel and surrounding sub-block is measured, therefore effectively remains the grain details of image.Such as
PPB (Probabilistic Patch-Based) method, this process employs SAR coherent speckle noise models and weight to maximize
Weight between possibility predication loop iteration similitude finally obtains suppression spot image after weighted average.But this method is in order to chase after
It asks and coherent speckle noise is inhibited as far as possible so that pressing down spot result is lost a large amount of grain details.
The prior art uses the thought of transform domain filtering and non-local mean filtering mostly, but is single use a kind of thought
Method due to the defect of itself be difficult effectively inhibit SAR coherent speckle noise while retain image grain details part.
Meanwhile occurring in similar removal image additive Gaussian noise field in conjunction with transform domain filtering and non-local mean
Three-dimensional bits matched filtering (Block-matching and 3D filtering, BM3D) algorithm of filter thought, the algorithm both existed
The grain details of image can also be effectively maintained while effectively removing image additive Gaussian noise, in removal image additive Gaussian
Field of noise has reached highest level, and occurs such method not yet in SAR Speckle reduction field, therefore the present invention exists
A kind of new SAR speckle suppression method is proposed on the basis of BM3D.
Summary of the invention
The purpose of the invention is to overcome above-mentioned the shortcomings of the prior art, a kind of SAR based on BM3D is proposed
Image speckle suppression method, by carrying out natural logrithm change to the image obtained after SAR image or the imaging of SAR initial data
Pretreatment after changing, so that the multiplying property coherent speckle noise in SAR image approaches additive Gaussian noise, then with BM3D algorithm to pre-
SAR image that treated carries out Speckle reduction, it is intended to while realization preferably inhibits coherent spot in SAR image, effectively protect
The grain details of image are stayed, step is implemented are as follows:
(1) natural logrithm transformation is carried out to L view SAR image:
SAR image I is regarded to the L of input0Or L is carried out by the initial data to SAR and regards SAR image I depending on obtained L is imaged0
Natural logrithm transformation is carried out, the log-domain SAR image I for regarding number as L is obtained1, L >=1；
(2) to log-domain SAR image I1It is pre-processed:
(2.1) to log-domain SAR image I1The biasing of coherent speckle noise mean value is carried out, obtaining coherent speckle noise mean value is zero
Log-domain SAR image I2；
(2.2) log-domain SAR image I is calculated2In each pixel pixel value maximum valueAnd pass throughTo I2
It is normalized, obtains the normalization log-domain SAR image I that value range is [0,1]3；
(3) normalization log-domain SAR image I is calculated3The standard deviation sigma of middle coherent speckle noise:
To normalization log-domain SAR image I3Wavelet transformation is carried out, low frequency sub-band wavelet coefficient LL, horizontal high-frequent are obtained
Band wavelet coefficient HL, vertical high frequency subband wavelet coefficient LH and diagonal high-frequency sub-band wavelet coefficient HH, and I is calculated by HH3In
Coherent speckle noise standard deviation sigma；
(4) using BM3D algorithm to normalization log-domain SAR image I3Carry out Speckle reduction:
(4.1) with log-domain SAR image I3Coordinate is that the point x of (0,0) is starting pixels point, and chooses around x k × k
Pixel forms image block P, k > 1, then will search in the region of n × n pixel composition around x similar with P
M image block Q forms similar block group G (P), n > k > 1, the expression formula of m >=1, G (P) are as follows:
G (P)={ Q:d (P, Q)≤τstep1}
Wherein, Euclidean distance of the d (P, Q) between image block P and image block Q similar with P, τstep1To measure similarity
Threshold value；
(4.2) similar block group G (P) is stacked into three-dimensional array G (P)3D, and to G (P)3DCollaboration hard -threshold filtering is carried out,
Obtain filter result
Wherein T3DhardAnd T-1 3DhardRespectively 3 D wavelet transformation and 3 D wavelet inverse transformation, γ are hard -threshold processing,λ3DFor threshold value；
(4.3) willM two dimensional image block group is reverted to, and is put back into G (P) in image I3In corresponding position,
Obtain the basic denoising result of G (P)It is right againAssembled, obtains image I3Basic denoising image Ibasic:
Wherein, xmFor m-th of similar block in similar block group G (P),For xmCharacteristic function, ωbasicFor xmPower
Weight:
Wherein, NharFor γ (T3Dhard(G (P))) in nonzero element number；
(4.4) to denoise image I substantiallybasicCoordinate is that the point x of (0,0) is starting pixels point, and chooses around x k × k
Pixel forms image block S, the l figure similar with P that then will be searched in the region of n × n pixel composition around x
As block T forms similar block group Gbasic(S), l >=1, and in image I3In same position take l image block composition similar block group
Gwie(S), Gbasic(S) expression formula are as follows:
Gbasic(S)={ Q:d (S, T)≤τstep2}
Wherein, Euclidean distance of the d (S, T) between image block S and image block T similar with S, τstep2To measure similarity
Threshold value；
(4.5) by similar block group Gwie(S) it is stacked into three-dimensional array Gwie(S)3D, to Gwie(S)3DCarry out collaboration wiener filter
Wave, obtaining filter result is
Wherein,WithRespectively 3 D wavelet transformation and 3 D wavelet inverse transformation, ωwieFor Wiener filtering coefficient:
(4.6) willL two dimensional image block group is reverted to, and is put back into Gwie(S) in image I3In corresponding position
It sets, obtains Gwie(S) basic denoising resultIt is right againAssembled, obtains image I3After Speckle reduction
Log-domain SAR image I4, formula are as follows:
Wherein, xlFor similar block group Gbasic(S) first of similar block in,It is xlCharacteristic function；
(5) to log-domain SAR image I4Carry out inverse transformation:
(5.1) to log-domain SAR image I4In each pixel pixel value multiplied byObtain renormalization SAR
Image I5；
(5.2) to renormalization SAR image I5Natural Exponents transformation is carried out, pointer field SAR image I is obtained6；
(5.3) by pointer field SAR image I6Middle pixel value is more than that the pixel value of 255 pixel is set as 255, obtains L
Depending on SAR image I0Suppression spot image Iresult。
Compared with the prior art, the invention has the following advantages:
The present invention is due to when obtaining the suppression spot image of former SAR image, first by carrying out natural logrithm to former SAR image
It is pre-processed after transformation, so that the multiplying property coherent speckle noise in former SAR image approaches additive Gaussian noise, then uses BM3D
Algorithm carries out Speckle reduction to pretreated SAR image, and the similar block search process in BM3D algorithm filters algorithm
When can utilize image autocorrelation, while to similar block group carry out 3 D wavelet transformation being capable of significantly more efficient burbling noise
With signal, avoid it is existing in the prior art can not achieve to coherent spot it is effective inhibit and grain details be effectively retained it is simultaneous
The defect of Gu, thus be effectively guaranteed SAR image it is subsequent classified, the accuracy of Target detection and identification.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention；
Fig. 2 is the simulation result diagram that the present invention and the prior art carry out Speckle reduction to SAR image；
Fig. 3 is the emulation that the present invention and the prior art carry out grain details reserving degree after Speckle reduction to SAR image
Result figure.
Specific embodiment
In the following with reference to the drawings and specific embodiments, present invention is further described in detail.
Referring to Fig.1, a kind of SAR image speckle suppression method based on BM3D, includes the following steps:
Step 1: natural logrithm transformation is carried out to L view SAR image:
(1.1) if there are initial data for the SAR image of input, L can be carried out to initial data and is regarded depending on obtained L is imaged
SAR image I0, the view number L of L >=1, imaging is bigger, then better, the of the invention phase of subsequent noise model Approximation effect of the invention
Dry spot suppression result is better, but the image I being imaged simultaneously0Resolution ratio is also lower, it is therefore desirable to be answered according to subsequent SAR image
L is arranged there is no next step, the present embodiment is directly carried out if initial data in the numerical value that L is chosen with situation, the SAR image if input
It is 4；
(1.2) SAR image I is regarded to L0Natural logrithm transformation is carried out, since the statistical model of SAR image coherent speckle noise is
Multiplicative noise model, if it is desired to Speckle reduction be carried out to SAR image using the thought of non-mean filter, must just pass through one
Multiplicative noise in image is become additive noise by fixed data map function, therefore selection operation of the present invention the most but has
The natural logrithm of effect is converted to image I0It is handled, obtains the log-domain SAR image I for regarding number as L1:
I1(i, j)=ln (I0(i, j)) i=0,1,2 ..., H j=0,1,2 ..., W
Wherein, I0(i, j) is image I0Middle abscissa is i, and ordinate is the pixel value of the point of j, and H is image I0Height, W
For image I0Width.
Step 2: to log-domain SAR image I1It is pre-processed:
(2.1) to log-domain SAR image I1Carry out the biasing of coherent speckle noise mean value:
(2.1.1) is due to obtained log-domain SAR image I1The mean value of middle coherent speckle noise is not zero, subsequent BM3D
In algorithm includes to the average weighted process of similar block group, if directly using BM3D algorithm to image I1Carry out coherent spot suppression
Meeting is made so that the suppression spot result finally obtained is not thorough, therefore the present invention is needed to log-domain SAR image I1In each pixel
The pixel value of point subtracts the mean value of log-domain coherent speckle noise corresponding with view number L, obtains image It:
It(i, j)=I1(i, j)-mean i=0,1,2 ..., H j=0,1,2 ..., W
Wherein, I1(i, j) is image I1Middle abscissa is i, and ordinate is the pixel value of the point of j, and H is image I1Height, W
For image I1Width, mean is that L regards the corresponding mean value of log-domain coherent speckle noise, and C is Euler's constant, size 0.577215；
(2.1.2) can make image I due to logarithmic transformation and mean value biasing processtIn certain points pixel value less than 0, and
Pixel value is no any physical significance less than 0, therefore by image ItIn all pixel values less than 0 be set as 0, obtain phase
The log-domain SAR image I that dry spot noise mean value is zero2:
I2(i, j)=relu (It(i, j)) i=0,1,2 ..., H j=0,1,2 ..., W
Wherein, relu () is line rectification function:
(2.2) log-domain SAR image I is calculated2In each pixel pixel value maximum valueAnd pass throughTo I2
It is normalized, obtains the normalization log-domain SAR image I that value range is [0,1]3:
Wherein,For log-domain SAR image I2In each pixel pixel value maximum value, I2(i, j) is image I2In
Abscissa is i, and ordinate is the pixel value of the point of j, and H is image I2Height, W be image I2Width；
Step 3: calculating normalization log-domain SAR image I3The standard deviation sigma of middle coherent speckle noise:
To normalization log-domain SAR image I3Wavelet transformation is carried out, low frequency sub-band wavelet coefficient LL, horizontal high-frequent are obtained
Band wavelet coefficient HL, vertical high frequency subband wavelet coefficient LH and diagonal high-frequency sub-band wavelet coefficient HH, and I is calculated by HH3In
Coherent speckle noise standard deviation sigma:
Wherein, median | HH | for the median of the absolute value of diagonal high-frequency sub-band wavelet coefficient；
Step 4: using BM3D algorithm to normalization log-domain SAR image I3Carry out Speckle reduction:
(4.1) with log-domain SAR image I3Coordinate is that the point x of (0,0) is starting pixels point, and chooses around x k × k
Pixel forms image block P, k > 1, then will search in the region of n × n pixel composition around x similar with P
M image block Q forms similar block group G (P), n > k > 1, the expression formula of m >=1, G (P) are as follows:
G (P)={ Q:d (P, Q)≤τstep1}
Wherein, Euclidean distance of the d (P, Q) between image block P and image block Q similar with P, τstep1To measure similarity
Threshold value；
(4.2) m similar block group G (P) is stacked into three-dimensional array G (P)3D, m >=1 does not have to stack when m is 1, only need
The third dimension of this similar block, which is set as 1, can be obtained G (P)3D, and to G (P)3DCollaboration hard -threshold filtering is carried out, is obtained
To filter result
Wherein T3DhardFor 3 D wavelet transformation, You Yici two-dimensional wavelet transformation and an one-dimensional wavelet transform composition, T-1 3DhardFor 3 D wavelet inverse transformation, it is made of primary one-dimensional inverse wavelet transform and a 2-d wavelet inverse transformation, γ is hard -threshold
Processing,λ3DFor threshold value；
Due to being the three-dimensional array G (P) being stacked into similar block group G (P) here3D3 D wavelet transformation is carried out, therefore can
When more effectively isolating noise and signal, then carry out hard -threshold filtering to the wavelet coefficient that 3 D wavelet transformation obtains
Effectively to filter out noise from signal, so that the present invention has the good ability for inhibiting SAR image coherent speckle noise.
(4.3) willM two dimensional image block group is reverted to, and is put back into G (P) in image I3In corresponding position,
Obtain the basic denoising result of G (P)It is right againAssembled, obtains image I3Basic denoising image Ibasic:
Wherein, xmFor m-th of similar block in similar block group G (P),For xmCharacteristic function, ωbasicFor xmPower
Weight:
Wherein, NharFor γ (T3Dhard(G (P))) in nonzero element number；
Due to being the basic denoising result to similar chunking hereIt is weighted and averaged, the non local filter utilized
The thought of wave, the existing autocorrelation of image itself enable the present invention effective when carrying out Speckle reduction to SAR image
Reservation image grain details；
(4.4) to denoise image I substantiallybasicCoordinate is that the point x of (0,0) is starting pixels point, and chooses around x k × k
Pixel forms image block S, the l figure similar with P that then will be searched in the region of n × n pixel composition around x
As block T forms similar block group Gbasic(S), l >=1, and in image I3In same position take l image block composition similar block group
Gwie(S), Gbasic(S) expression formula are as follows:
Gbasic(S)={ Q:d (S, T)≤τstep2}
Wherein, Euclidean distance of the d (S, T) between image block S and image block T similar with S, τstep2To measure similarity
Threshold value；
(4.5) by similar block group Gwie(S) it is stacked into three-dimensional array Gwie(S)3D, to Gwie(S)3DCarry out collaboration wiener filter
Wave, obtaining filter result is
Wherein,WithRespectively 3 D wavelet transformation and 3 D wavelet inverse transformation, ωwieFor Wiener filtering coefficient:
(4.6) willL two dimensional image block group is reverted to, and is put back into Gwie(S) in image I3In corresponding position
It sets, obtains Gwie(S) basic denoising resultIt is right againAssembled, obtains image I3After Speckle reduction
Log-domain SAR image I4, formula are as follows:
Wherein, xlFor similar block group Gbasic(S) first of similar block in,It is xlCharacteristic function；
Step 5: to log-domain SAR image I4Carry out inverse transformation:
(5.1) to log-domain SAR image I4In each pixel pixel value multiplied byObtain renormalization SAR
Image I5；
(5.2) to renormalization SAR image I5Natural Exponents transformation is carried out, pointer field SAR image I is obtained6, transformation for mula
Are as follows:
Wherein e is natural constant, size 2.71828, I5(i, j) is image I5Middle abscissa is i, and ordinate is the point of j
Pixel value, H be image I5Height, W be image I5Width；
(5.3) by pointer field SAR image I6Middle pixel value is more than that the pixel value of 255 pixel is set as 255, obtains L
Depending on SAR image I0Suppression spot image Iresult。
Below in conjunction with emulation experiment, technical effect of the invention is described further:
1. simulated conditions and content:
Simulated conditions: simulation software Intel (R) Core (TM) i7-7700HQ of dominant frequency 2.8GHZ, memory 8GB it is hard
It is carried out under the software environment of part environment and MATLAB R2017b.
Emulation content: coherent spot is carried out to the SAR image obtained after the imaging of SAR initial data with the present invention and the prior art
Inhibit.
2. analysis of simulation result:
(2.1) SAR image Speckle reduction field is used to measure the suppression spot figure that the index of Speckle reduction degree is
As the equivalent number of flat site, i.e. ENL value, corresponding ENL value is higher, indicates that the suppression spot ability of this method is stronger.Its result
As shown in Fig. 2 and table one:
Fig. 2 (a) is that initial data is imaged with range Doppler algorithm (range Doppler algorithm, RDA)
To the single-view image of Vancouver area in 2002, the flat site that the white edge in image is chosen is used to calculate ENL, Fig. 2 (b), Fig. 2
(c) and Fig. 2 (d) is respectively method ST, and method PPB and the present invention carry out the result figure of Speckle reduction to SAR image respectively, from
It rough can find out that the suppression result of the invention in flat site coherent speckle noise is the most apparent in result figure.Table 1 is
The ENL value calculated with the corresponding suppression spot image of original SAR image and three kinds of methods.
| Method | Original SAR image | ST | PPB | The present invention |
| ENL | 0.43 | 4.63 | 8.47 | 10.28 |
Table 1
As shown in Fig. 2 and table 1, the ENL value for the suppression spot image that the present invention obtains is maximum, shows the present invention to SAR image phase
The rejection ability of dry spot noise is better than the prior art.
(2.2) SAR image Speckle reduction field is used to measure after Speckle reduction to image texture details reserving degree
Index be original SAR image and suppression spot image ratio images, the texture structure that corresponding ratio images contain is fewer, explanation
The suppression spot method is higher to the reserving degree of image texture details.Its result is as shown in Figure 3:
Fig. 3 (a), Fig. 3 (b) and Fig. 3 (c) are respectively method ST, the ratio images that method PPB and the present invention obtain, by Fig. 2
(a) it carries out point pixel-by-pixel with Fig. 2 (b), Fig. 2 (c) and Fig. 2 (d) respectively to be divided by obtain, it can be seen that the ratio figure that the present invention obtains
Picture Fig. 3 (c) is substantially free of texture structure, therefore the present invention protects the grain details of image after to SAR image Speckle reduction
That stays is best.
By the above experimental program, demonstrating the present invention both has stronger Speckle noise removal ability, while also can
Enough it is effectively maintained the grain details of image.
The invention proposes a kind of SAR image speckle suppression methods based on BM3D, by SAR image or SAR
The image obtained after initial data imaging pre-process after taking logarithmic transformation, so that the multiplying property coherent spot of original SAR image
Noise approaches gaussian additive noise, and image oneself is utilized when the similar block search process in BM3D algorithm enables algorithm to filter
Correlation, to similar block group carry out 3 D wavelet transformation can significantly more efficient burbling noise and signal so that the present invention is having
Effect inhibits the grain details part that can be good at retaining image while image coherent speckle noise, to be effectively guaranteed
SAR image is subsequent to be classified, the accuracy of Target detection and identification.
Claims (6)
1. a kind of SAR image speckle suppression method based on BM3D, which comprises the steps of:
(1) natural logrithm transformation is carried out to L view SAR image:
SAR image I is regarded to the L of input0Or L is carried out by the initial data to SAR and regards SAR image I depending on obtained L is imaged0It carries out
Natural logrithm transformation, obtains the log-domain SAR image I for regarding number as L1, L >=1；
(2) to log-domain SAR image I1It is pre-processed:
(2.1) to log-domain SAR image I1The biasing of coherent speckle noise mean value is carried out, the logarithm that coherent speckle noise mean value is zero is obtained
Domain SAR image I2；
(2.2) log-domain SAR image I is calculated2In each pixel pixel value maximum valueAnd pass throughTo I2Returned
One changes, and obtains the normalization log-domain SAR image I that value range is [0,1]3；
(3) normalization log-domain SAR image I is calculated3The standard deviation sigma of middle coherent speckle noise:
To normalization log-domain SAR image I3Wavelet transformation is carried out, it is small to obtain low frequency sub-band wavelet coefficient LL, horizontal high-frequent subband
Wave system number HL, vertical high frequency subband wavelet coefficient LH and diagonal high-frequency sub-band wavelet coefficient HH, and I is calculated by HH3In phase
The standard deviation sigma of dry spot noise；
(4) using BM3D algorithm to normalization log-domain SAR image I3Carry out Speckle reduction:
(4.1) with log-domain SAR image I3Coordinate is that the point x of (0,0) is starting pixels point, and chooses k × k pixel around x
Image block P, k > 1 are formed, the m figure similar with P that then will be searched in the region of n × n pixel composition around x
As block Q composition similar block group G (P), n > k > 1, the expression formula of m >=1, G (P) are as follows:
G (P)={ Q:d (P, Q)≤τstep1}
Wherein, Euclidean distance of the d (P, Q) between image block P and image block Q similar with P, τstep1For the threshold for measuring similarity
Value；
(4.2) similar block group G (P) is stacked into three-dimensional array G (P)3D, and to G (P)3DCollaboration hard -threshold filtering is carried out, is filtered
Wave result
Wherein T3DhardAnd T-1 3DhardRespectively 3 D wavelet transformation and 3 D wavelet inverse transformation, γ are hard -threshold processing,λ3DFor threshold value；
(4.3) willM two dimensional image block group is reverted to, and is put back into G (P) in image I3In corresponding position, obtain G
(P) basic denoising resultIt is right againAssembled, obtains image I3Basic denoising image Ibasic:
Wherein, xmFor m-th of similar block in similar block group G (P),For xmCharacteristic function, ωbasicFor xmWeight:
Wherein, NharFor γ (T3Dhard(G (P))) in nonzero element number；
(4.4) to denoise image I substantiallybasicCoordinate is that the point x of (0,0) is starting pixels point, and chooses k × k pixel around x
Point composition image block S, the l image block similar with P that then will be searched in the region of n × n pixel composition around x
T forms similar block group Gbasic(S), l >=1, and in image I3In same position take l image block composition similar block group Gwie
(S), Gbasic(S) expression formula are as follows:
Gbasic(S)={ Q:d (S, T)≤τstep2}
Wherein, Euclidean distance of the d (S, T) between image block S and image block T similar with S, τstep2For the threshold for measuring similarity
Value；
(4.5) by similar block group Gwie(S) it is stacked into three-dimensional array Gwie(S)3D, to Gwie(S)3DCollaboration Wiener filtering is carried out, is obtained
Filter result is
Wherein,WithRespectively 3 D wavelet transformation and 3 D wavelet inverse transformation, ωwieFor Wiener filtering coefficient:
(4.6) willL two dimensional image block group is reverted to, and is put back into Gwie(S) in image I3In corresponding position,
Obtain Gwie(S) basic denoising resultIt is right againAssembled, obtains image I3Pair after Speckle reduction
Number field SAR image I4, formula are as follows:
Wherein, xlFor similar block group Gbasic(S) first of similar block in, χxlIt is xlCharacteristic function；
(5) to log-domain SAR image I4Carry out inverse transformation:
(5.1) to log-domain SAR image I4In each pixel pixel value multiplied by maxI2, obtain renormalization SAR image
I5；
(5.2) to renormalization SAR image I5Natural Exponents transformation is carried out, pointer field SAR image I is obtained6；
(5.3) by pointer field SAR image I6Middle pixel value is more than that the pixel value of 255 pixel is set as 255, obtains L view SAR
Image I0Suppression spot image Iresult。
2. the SAR image speckle suppression method according to claim 1 based on BM3D, which is characterized in that step (1) institute
That states carries out natural logrithm transformation, transformation for mula to L view SAR image are as follows:
I1(i, j)=ln (I0(i, j)) i=0,1,2 ..., H j=0,1,2 ..., W
Wherein, I0(i, j) is image I0Middle abscissa is i, and ordinate is the pixel value of the point of j, and H is image I0Height, W be figure
As I0Width.
3. the SAR image speckle suppression method according to claim 1 based on BM3D, it is characterised in that: step (2.1)
Described in log-domain SAR image I1The biasing of coherent speckle noise mean value is carried out, realizes that steps are as follows:
(2.1.1) is to log-domain SAR image I1In each pixel pixel value subtract it is relevant with the corresponding log-domain of view number L
The mean value of spot noise obtains image It:
It(i, j)=I1(i, j)-mean i=0,1,2 ..., H j=0,1,2 ..., W
Wherein, I1(i, j) is image I1Middle abscissa is i, and ordinate is the pixel value of the point of j, and H is image I1Height, W be figure
As I1Width, mean is that L regards the corresponding mean value of log-domain coherent speckle noise, and C is Euler's constant, size 0.577215；
(2.1.2) is by image ItIn all pixel values less than 0 be set as 0, obtain the log-domain that coherent speckle noise mean value is zero
SAR image I2:
I2(i, j)=relu (It(i, j)) i=0,1,2 ..., H j=0,1,2 ..., W
Wherein, relu () is line rectification function:
4. the SAR image speckle suppression method according to claim 1 based on BM3D, it is characterised in that: step (2.2)
Described in log-domain SAR image I2It is normalized, specific formula are as follows:
Wherein,For log-domain SAR image I2In each pixel pixel value maximum value, I2(i, j) is image I2Middle horizontal seat
It is designated as i, ordinate is the pixel value of the point of j, and H is image I2Height, W be image I2Width.
5. the SAR image speckle suppression method according to claim 1 based on BM3D, it is characterised in that: in step (3)
The calculating I3In coherent speckle noise standard deviation sigma, calculation formula are as follows:
Wherein, median | HH | for the median of the absolute value of diagonal high-frequency sub-band wavelet coefficient.
6. the SAR image speckle suppression method according to claim 1 based on BM3D, it is characterised in that: step (5.2)
To renormalization SAR image I5Natural Exponents transformation is carried out, pointer field SAR image I is obtained6, transformation for mula are as follows:
Wherein e is natural constant, size 2.71828, I5(i, j) is image I5Middle abscissa is i, and ordinate is the picture of the point of j
Element value, H are image I5Height, W be image I5Width.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910162471.5A CN109919870B (en) | 2019-03-05 | 2019-03-05 | SAR image speckle suppression method based on BM3D |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910162471.5A CN109919870B (en) | 2019-03-05 | 2019-03-05 | SAR image speckle suppression method based on BM3D |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109919870A true CN109919870A (en) | 2019-06-21 |
| CN109919870B CN109919870B (en) | 2021-04-02 |
Family
ID=66963277
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910162471.5A Active CN109919870B (en) | 2019-03-05 | 2019-03-05 | SAR image speckle suppression method based on BM3D |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919870B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340702A (en) * | 2020-02-24 | 2020-06-26 | 江南大学 | Sparse reconstruction method for high-frequency ultrasonic microscopic imaging of tiny defects based on blind estimation |
| CN111461999A (en) * | 2020-03-13 | 2020-07-28 | 西安工程大学 | SAR image speckle suppression method based on super-pixel similarity measurement |
| CN111861905A (en) * | 2020-06-17 | 2020-10-30 | 浙江工业大学 | SAR image speckle noise suppression method based on Gamma-Lee filtering |
| CN112633298A (en) * | 2020-12-28 | 2021-04-09 | 深圳大学 | Method for measuring similarity of image/image block |
| CN112734666A (en) * | 2020-12-31 | 2021-04-30 | 西安电子科技大学 | SAR image speckle non-local mean suppression method based on similarity value |
| CN112927165A (en) * | 2021-03-22 | 2021-06-08 | 重庆邮电大学 | SAR image speckle suppression method based on NSST domain three-dimensional block matching |
| CN114037616A (en) * | 2021-09-22 | 2022-02-11 | 南京莱斯电子设备有限公司 | SAR image noise suppression method and equipment |
| CN114627148A (en) * | 2021-07-05 | 2022-06-14 | 广东省科学院广州地理研究所 | Coastal zone aquaculture water body object extraction method and device based on microwave remote sensing |
| CN115031651A (en) * | 2022-06-07 | 2022-09-09 | 天津大学 | Improved BM3D denoising OFDR distributed strain measurement method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020922A (en) * | 2013-01-10 | 2013-04-03 | 西安电子科技大学 | PCA (principal component analysis) transformation based SAR (synthetic aperture radar) image speckle suppression method |
| CN103871031A (en) * | 2014-03-07 | 2014-06-18 | 西安电子科技大学 | Kernel regression-based SAR image coherent speckle restraining method |
| CN105913083A (en) * | 2016-04-08 | 2016-08-31 | 西安电子科技大学 | Dense SAR-SIFT and sparse coding-based SAR classification method |
| CN106971382A (en) * | 2017-03-16 | 2017-07-21 | 中国人民解放军国防科学技术大学 | A kind of SAR image speckle suppression method |
| CN109389101A (en) * | 2018-11-02 | 2019-02-26 | 电子科技大学 | A kind of SAR image target recognition method based on denoising autoencoder network |
-
2019
- 2019-03-05 CN CN201910162471.5A patent/CN109919870B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020922A (en) * | 2013-01-10 | 2013-04-03 | 西安电子科技大学 | PCA (principal component analysis) transformation based SAR (synthetic aperture radar) image speckle suppression method |
| CN103871031A (en) * | 2014-03-07 | 2014-06-18 | 西安电子科技大学 | Kernel regression-based SAR image coherent speckle restraining method |
| CN105913083A (en) * | 2016-04-08 | 2016-08-31 | 西安电子科技大学 | Dense SAR-SIFT and sparse coding-based SAR classification method |
| CN106971382A (en) * | 2017-03-16 | 2017-07-21 | 中国人民解放军国防科学技术大学 | A kind of SAR image speckle suppression method |
| CN109389101A (en) * | 2018-11-02 | 2019-02-26 | 电子科技大学 | A kind of SAR image target recognition method based on denoising autoencoder network |
Non-Patent Citations (2)
| Title |
|---|
| MURALIMOHANBABU.YA,ET AL.: "A modified BM3D algorithm for SAR image despeckling", 《4THINTERNATIONAL CONFERENCE ON ECO-FRIENDLY COMPUTING AND COMMUNICATION SYSTEMS》 * |
| 鲁自立 等: "多方向高斯窗匹配模板的SAＲ－BM3D相干斑抑制算法", 《无线电工程》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340702A (en) * | 2020-02-24 | 2020-06-26 | 江南大学 | Sparse reconstruction method for high-frequency ultrasonic microscopic imaging of tiny defects based on blind estimation |
| CN111461999A (en) * | 2020-03-13 | 2020-07-28 | 西安工程大学 | SAR image speckle suppression method based on super-pixel similarity measurement |
| CN111461999B (en) * | 2020-03-13 | 2023-02-14 | 西安工程大学 | SAR image speckle suppression method based on super-pixel similarity measurement |
| CN111861905A (en) * | 2020-06-17 | 2020-10-30 | 浙江工业大学 | SAR image speckle noise suppression method based on Gamma-Lee filtering |
| CN111861905B (en) * | 2020-06-17 | 2022-07-15 | 浙江工业大学 | SAR image speckle noise suppression method based on Gamma-Lee filtering |
| CN112633298B (en) * | 2020-12-28 | 2023-07-18 | 深圳大学 | Method for measuring similarity of image/image block |
| CN112633298A (en) * | 2020-12-28 | 2021-04-09 | 深圳大学 | Method for measuring similarity of image/image block |
| CN112734666A (en) * | 2020-12-31 | 2021-04-30 | 西安电子科技大学 | SAR image speckle non-local mean suppression method based on similarity value |
| CN112927165A (en) * | 2021-03-22 | 2021-06-08 | 重庆邮电大学 | SAR image speckle suppression method based on NSST domain three-dimensional block matching |
| CN114627148A (en) * | 2021-07-05 | 2022-06-14 | 广东省科学院广州地理研究所 | Coastal zone aquaculture water body object extraction method and device based on microwave remote sensing |
| CN114627148B (en) * | 2021-07-05 | 2024-01-02 | 广东省科学院广州地理研究所 | Method and device for extracting coastal zone aquaculture water body object based on microwave remote sensing |
| CN114037616A (en) * | 2021-09-22 | 2022-02-11 | 南京莱斯电子设备有限公司 | SAR image noise suppression method and equipment |
| CN114037616B (en) * | 2021-09-22 | 2024-04-05 | 南京莱斯电子设备有限公司 | SAR image noise suppression method and device |
| CN115031651A (en) * | 2022-06-07 | 2022-09-09 | 天津大学 | Improved BM3D denoising OFDR distributed strain measurement method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109919870B (en) | 2021-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919870A (en) | A kind of SAR image speckle suppression method based on BM3D | |
| CN101847257B (en) | Image denoising method based on non-local means and multi-level directional images | |
| JP6057881B2 (en) | Method for removing noise from input image consisting of pixels containing noise | |
| CN103279957B (en) | A kind of remote sensing images area-of-interest exacting method based on multi-scale feature fusion | |
| CN103020922A (en) | PCA (principal component analysis) transformation based SAR (synthetic aperture radar) image speckle suppression method | |
| CN109961408B (en) | Photon counting image denoising method based on NSCT and block matching filtering | |
| CN108428221A (en) | A kind of neighborhood bivariate shrinkage function denoising method based on shearlet transformation | |
| CN103077507A (en) | Beta algorithm-based multiscale SAR (Synthetic Aperture Radar) image denoising method | |
| CN110060220A (en) | Based on the image de-noising method and system for improving BM3D algorithm | |
| Devapal et al. | Discontinuity adaptive SAR image despeckling using curvelet-based BM3D technique | |
| CN103426145A (en) | Synthetic aperture sonar speckle noise suppression method based on multiresolution analysis | |
| CN102722878A (en) | SAR (synthetic aperture radar) image despeckle method based on target extraction and PPB (probabilistic patch-based filter) algorithm | |
| CN109242797B (en) | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion | |
| CN102903083A (en) | Context model and dual-tree complex wavelet transform-based denoising method for underwater sonar image | |
| CN111461999A (en) | SAR image speckle suppression method based on super-pixel similarity measurement | |
| CN105096274A (en) | Infrared image noise reduction method based on non-subsampled contourlet domain mixed statistical model | |
| Sun et al. | Overview of SAR image denoising based on transform domain | |
| Zhang et al. | Bayesian-based speckle suppression for SAR image using contourlet transform | |
| CN112927169A (en) | Remote sensing image denoising method based on wavelet transformation and improved weighted nuclear norm minimization | |
| Bhargava et al. | An Effective Method for Image Denoising Using Non-local Means and Statistics based Guided Filter in Nonsubsampled Contourlet Domain. | |
| Tang et al. | How SAR Image Denoise Affects the Performance of DCNN-Based Target Recognition Method | |
| Liu et al. | An improved synthetic aperture radar image denoising method based on block-matching and 3D filtering | |
| CN112184574B (en) | Method and system for eliminating speckle noise of remote sensing image | |
| Bhonslei | Additive and multiplicative noise removal from medical images using bivariate thresholding by dual tree complex wavelet transform | |
| CN103839240A (en) | SAR image despeckle method based on probability model and NSDT |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |