CN109035152B - Non-local mean filtering method for synthetic aperture radar image - Google Patents
Non-local mean filtering method for synthetic aperture radar image Download PDFInfo
- Publication number
- CN109035152B CN109035152B CN201810502002.9A CN201810502002A CN109035152B CN 109035152 B CN109035152 B CN 109035152B CN 201810502002 A CN201810502002 A CN 201810502002A CN 109035152 B CN109035152 B CN 109035152B
- Authority
- CN
- China
- Prior art keywords
- image
- value
- coefficient
- noise
- image block
- 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
- 238000001914 filtration Methods 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000015556 catabolic process Effects 0.000 claims abstract description 32
- 238000006731 degradation reaction Methods 0.000 claims abstract description 32
- 230000001427 coherent effect Effects 0.000 claims abstract description 11
- 238000009826 distribution Methods 0.000 claims description 17
- 238000011524 similarity measure Methods 0.000 claims description 11
- 230000003044 adaptive effect Effects 0.000 claims description 8
- 238000005315 distribution function Methods 0.000 claims description 6
- 238000003384 imaging method Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000005259 measurement Methods 0.000 abstract description 13
- 238000012545 processing Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 description 9
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a synthetic aperture radar image non-local mean filtering method, and belongs to the field of radar image processing. The invention introduces the heterogeneity measure into the non-local mean algorithm, and improves the weight measurement mode of the non-local mean algorithm. Firstly, aiming at the traditional non-local mean method, an equal weight value weighting method is adopted for calculating distance measure, a variation coefficient is introduced to calculate the weighted weight value of the distance measure of a pixel point in an image block window, and the variation coefficient can represent the texture direction of an image, so that the similarity of the image block after the introduction of the variation coefficient and weighting can more effectively capture the image block with similarity in the texture and direction, thereby avoiding the ambiguity of marginality and directivity; then, a self-adaptive degradation parameter function based on the variation coefficient is designed, the degradation parameter is adjusted in a self-adaptive mode, and the heterogeneous region can be effectively protected from being excessively smooth. Therefore, the method can more effectively keep the texture edge information while filtering out the coherent spots.
Description
Technical Field
The invention belongs to the field of radar image processing, and particularly relates to a synthetic aperture radar image non-local mean filtering method.
Background
Synthetic Aperture Radar (SAR) is an active high-resolution imaging sensor, can perform observation all day long and all weather, has strong penetrability, and is widely applied to the fields of military investigation, disaster monitoring, surface covering layer detection and the like. However, since the SAR imaging is performed by using echo coherent superposition, speckle noise inevitably exists in the SAR image, and the speckle noise reduces the visual effect of the SAR image, thereby causing difficulty in subsequent image segmentation and target identification. Therefore, speckle suppression of the SAR image is of great significance to segmentation, classification and identification of the SAR image.
The speckle filtering method of the SAR image can be divided into spatial filtering and transform domain filtering. The spatial filtering means that neighborhood pixels of pixels to be filtered are weighted and averaged to obtain a filtered gray value, and the methods are represented by Lee filtering, Frost filtering, Gamma-Map filtering and the like, which are provided based on a specific speckle distribution model, so that the method only has a good filtering effect on SAR images with specific distribution. Transform domain filtering refers to a method of transforming an image from a time domain to a corresponding transform domain by fourier transform or wavelet transform, then performing filtering processing on the transform domain, and finally performing inverse transform to return a time domain two-dimensional image. Both of these methods are essentially based on local or semi-local filtering methods, without taking into account the structural redundancy information of the image itself.
In order to fully utilize the self structural redundancy information of an image, a Non-Local Means (NLM) algorithm is provided, the algorithm utilizes the structural redundancy of the image, an image block similar to a target image block is searched in the image, corresponding weight is given to a pixel point according to the similarity between the image blocks, and an estimated value of the target pixel point is obtained by a weighted average method. In the prior art, a method is provided for providing a BM3D filtering method based on NLM algorithm by combining wavelet domain filtering and wiener filtering ideas, the method includes two estimations of basic estimation and final estimation, the preliminary estimation gathers similar blocks into a three-dimensional array, filtering the three-dimensional array by adopting a joint hard threshold, returning filtered image blocks to their original positions to obtain overlapped block estimation, then carrying out weighted average to obtain a preliminarily estimated image, the final estimation is to gather a noisy image and the preliminarily estimated image into two three-dimensional images respectively by utilizing block similarity, then using joint wiener filtering, putting the filtered image blocks to the original positions to obtain an overlapped estimated image, weighting to obtain a final estimated image, and obtaining a good effect. However, the above methods are only applicable to additive white gaussian noise, but not on SAR images with multiplicative noise, and therefore, many scholars have studied the application of NLM algorithm on SAR images. Therefore, a (PPB) non-local mean filtering method based on image block probability distribution is provided, the method utilizes the noise distribution characteristic of the image, maximum likelihood estimation is adopted when the block similarity is calculated, the estimation is more and more accurate in the iteration process, and the method obtains good filtering effect on the SAR image; in addition, another method proves that the ratio measure can represent the distance measure of the SAR image more stably, and provides a homogeneous region non-local mean filtering method based on the ratio measure, but the method is only suitable for SAR image denoising in the homogeneous region; on the basis of BM3D, an SAR-BM3D method is provided, the method for improving block similarity measurement is provided, the problem of wavelet coefficient shrinkage is solved, and a very good filtering effect is obtained on an SAR image; and a method is also provided, on the basis of the ratio measurement non-local mean filtering, the window weight of the distance measurement is improved, so that the ratio measurement non-local mean filtering proposed by Hongxiao Feng can be popularized to a heterogeneous region. However, when the distance measurement is calculated by these methods, the average value of the gray-scale distances of the pixels in the window or the gaussian weighted value is used, and the texture direction information of the image is not considered, but a large amount of texture direction information exists in the SAR image, and the information should be an important basis for the distance measurement. In addition, in the NLM algorithm, the weight is a gaussian kernel function depending on the distance measure, where the degradation parameter has a very important influence on the filtering effect, however, there are few documents on the research on the degradation parameter, which is generally set according to an empirical value, and the adaptive denoising effect is not achieved.
Disclosure of Invention
In order to solve the above problems, the invention provides a non-local mean filtering method for a synthetic aperture radar image, which introduces a Coefficient of Variation (CV) to construct an anisotropic two-dimensional weighting window that changes adaptively according to the image; in addition, a method for adaptively adjusting degradation parameters based on the variation coefficient is provided, so that the image has a good denoising effect in a homogeneous region, and meanwhile, edge texture details can be kept in a heterogeneous region.
A synthetic aperture radar image non-local mean filtering method comprises the following steps:
step 1, obtaining a synthetic aperture radar amplitude image and initializing parameters, wherein the initialized parameters comprise a search window size Ds, an image window size Ds, a degradation parameter coefficient gamma and an S curve gradient parameter xi;
step 2, calculating to obtain speckle distribution of the image, and obtaining noise variance of the speckle based on the speckle distribution;
step 3, calculating an image variation coefficient according to the standard deviation and the mean value of the image;
step 4, calculating self-adaptive degradation parameters according to the image variation coefficient;
step 5, constructing a self-adaptive filtering window based on the variation coefficient;
and 6, obtaining the value of the pixel point based on the non-local algorithm of the heterogeneity measure.
Further, the step 1 further includes the following steps:
setting a pixel point i as a pixel positioned at an image position i, setting an image block i as an image block with the size ds and the pixel point i as the center, and searching a region omegaiThe image blocks are image blocks with the size Ds and with the pixel point i as the center.
Further, the step 2 comprises the following steps:
when the image pixel spacing is comparable to the radar imaging resolution unit, the noise power is uncorrelated, and the speckle is incoherent multiplicative noise, denoted as
υ＝u·n
Wherein upsilon is an observed value, u is a radar Reflection Coefficient (RCS), n is speckle noise, and the equivalent visual number of the image can be obtained according to the standard deviation and the mean value of the image
Wherein, L is equivalent vision, sigma is standard deviation of the image, mu is mean value of the image, the coherent speckles are completely developed, and the distribution function of the intensity condition is
Wherein Gamma (·) is Gamma function, radar reflection coefficient is set to 1, and completely developed coherent speckle noise distribution is obtained
The variance of the noise intensity is calculated as
The amplitude value of the image is the square root of the intensity, and the amplitude condition distribution of the obtained observation value is
Setting the radar reflection coefficient to 1, the distribution function of amplitude noise can be obtained
The noise variance is calculated as
Further, the step 3 includes the following steps:
coefficient of variation of
Wherein, CV is a local variation coefficient of the pixel point.
Further, the step 4 includes the following steps:
adaptively adjusting the coefficient of variation based on an S-curve represented as
Wherein the parameter tau is used for controlling the offset of the S curve, and the parameter xi is used for controlling the inclination of the S curve;
according to the variation coefficient, for each pixel point in the image, the self-adaptive degradation parameter is expressed as
Wherein h' is an adaptive degradation factor, h ═ γ σ, CVmaxIs the maximum local variation coefficient of the entire image.
Further, the step 5 comprises the following steps:
according to the multiplicative noise model in the step 2, the noise is white Gaussian noise with the mean value as a unit value, and the variation coefficient of one image block is expressed as
Where Ψ is an image block in the observation image v, and E (u) if the image block is in a homogeneous regionΨ 2)＝[E(uΨ)]2The coefficient of variation can be simplified to
Based on the noise value as a unit value
The image block variation coefficient in the homogeneous region is approximate to the image noise variance, and the anisotropic filtering window constructed according to the local variation coefficient of the image is expressed as
ACV(i,k)＝exp(-|CV(i,k)-CV(i)|2)
Wherein A isCVAnd (i, k) is an anisotropic weighting window taking i as a center, and CV (i, k) is a variation coefficient of the kth pixel point in the image block taking i as a center.
Further, the step 6 includes the following steps:
step 61, calculating image block similarity measure
Based on a ratio measure and the adaptive filtering window, a similarity measure is expressed as
Wherein d isCV(i,j)Is the similarity measure of the image blocks with the pixel points i and j as the center, psi is the image block, ZCV(i) The function xi (x) is a normalization factor of the coefficients of variation in the image block, and has an expression xi (x) x +1/x, ui(k) Is the gray value, u, of the k-th pixel point of the image block centered at ij(k) The gray value of the kth pixel point of the image block taking j as the center;
step 62, calculating the weight of the pixel point
According to the NLM algorithm, based on the Gaussian kernel function, calculating the weight between the pixel points
Wherein,
is a normalized coefficient;
step 63, calculating pixel values of pixel points
According to the NLM algorithm, for a pixel point i, a search area omega is formediThe pixel points in the image are overlapped according to the weight value to obtain the estimated value of the pixel point i
Wherein u (j) is the gray value corresponding to the pixel j.
The invention has the beneficial effects that: the invention introduces the heterogeneity measure into the non-local mean algorithm, and improves the weight measurement mode of the non-local mean algorithm. Firstly, aiming at the traditional non-local mean method, an equal weight value weighting method is adopted for calculating distance measure, a variation coefficient is introduced to calculate the weighted weight value of the distance measure of a pixel point in an image block window, and the variation coefficient can represent the texture direction of an image, so that the similarity of the image block after the introduction of the variation coefficient and weighting can more effectively capture the image block with similarity in the texture and direction, thereby avoiding the ambiguity of marginality and directivity; then, a self-adaptive degradation parameter function based on the variation coefficient is designed, the degradation parameter is adjusted in a self-adaptive mode, and the heterogeneous region can be effectively protected from being excessively smooth. Therefore, the method can more effectively keep the texture edge information while filtering out the coherent spots.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a flowchart of step 6 in fig. 1.
Detailed Description
The embodiments of the present invention will be further described with reference to the accompanying drawings.
Referring to fig. 1, the non-local mean filtering method for synthetic aperture radar image according to the present invention is implemented by the following steps:
step 1, obtaining a synthetic aperture radar amplitude image and initializing parameters, wherein the initialized parameters comprise a search window size Ds, an image window size Ds, a degradation parameter coefficient gamma and an S curve gradient parameter xi.
In this embodiment, an input image SAR airborne amplitude map includes homogeneous region image blocks a and heterogeneous region image blocks B, the size of the image blocks is 30 × 30, the image blocks are used for image filtering result evaluation, the wave band is an X wave band, the resolution is 3 meters, and the size of the image is 256 × 256. The initialized parameters comprise a search window size Ds, an image window size Ds, a degradation parameter coefficient gamma and an S curve gradient parameter xi.
In this embodiment, a pixel point i is set as a pixel located at an image position i, an image block i is an image block with a size ds and the pixel point i as a center, and a search area ΩiThe image blocks are image blocks with the size Ds and with the pixel point i as the center.
And 2, calculating to obtain the speckle distribution of the image, and obtaining the noise variance of the speckle based on the speckle distribution.
In this embodiment, step 2 is implemented by the following process:
when the image pixel spacing is comparable to the radar imaging resolution unit, the noise power is uncorrelated, and the speckle can be considered to be incoherent multiplicative noise, denoted as
υ＝u·n
Wherein upsilon is an observed value, u is a real radar Reflection Coefficient (RCS), n is speckle noise, an equivalent view of the image can be obtained according to a standard deviation and a mean value of the input image, and an expression of the equivalent view is
Wherein, L is equivalent vision, sigma is standard deviation of the image, mu is mean value of the image, for SAR image with vision of L, several spots are completely developed, the intensity condition distribution function can be expressed as
Wherein Gamma (·) is Gamma function, radar reflection coefficient is set to 1, and completely developed coherent speckle noise distribution is obtained
The variance of the noise intensity is calculated as
The amplitude value of the SAR image is the square root of the intensity, and the amplitude condition distribution of the obtained observed value is
Setting the radar reflection coefficient to 1, the distribution function of amplitude noise can be obtained
The noise variance is calculated as
And 3, calculating an image variation coefficient according to the standard deviation and the mean value of the image.
In this embodiment, step 3 is implemented by the following process:
the coefficient of variation is the ratio of the standard deviation and the mean of the image, i.e.
For each pixel in the SAR image, the local variation coefficient of the pixel position is the variation coefficient of the pixel estimated by taking the local neighborhood image block of the pixel, the local neighborhood size of the pixel is consistent with the image block size definition, and the ds multiplied by ds image block taking the pixel as the center is taken to be used for calculating the variation coefficient of the center pixel:
wherein σΨAnd muΨAnd the standard deviation and the mean value of the local neighborhood of the pixel point.
And 4, calculating a self-adaptive degradation parameter according to the image variation coefficient.
In this embodiment, step 4 is implemented by the following process:
in the traditional NLM algorithm, the degradation factor of the exponential function of the weight is calculated according to the similarity measure and is constant and unchangeable in the whole image, so the selection of the degradation factor is very important, if the degradation factor is too large, the excessive smoothness is caused, the texture edge of the image is not favorably maintained, if the degradation factor is too small, a satisfactory filtering effect cannot be achieved in a homogeneous region, and because the degradation factor can distinguish the homogeneous region from a heterogeneous region, the invention provides a self-adaptive degradation factor calculation method by combining the degradation factor, so that the degradation factor can be automatically adjusted to adapt to the filtering requirements of images in different regions. The definition of the variation coefficient shows that the variation coefficient of the homogeneous region is smaller and close to the noise variance, the variation coefficient of the heterogeneous region is far larger than the noise variance, and by combining the analysis of the degradation factor, the variation coefficient is adaptively adjusted by using an S-shaped curve, and the expression of the S-shaped curve is as follows:
wherein, the parameter tau is used for controlling the offset of the S curve, and the parameter xi is used for controlling the inclination of the S curve.
According to the variation coefficient, for each pixel point in the image, the adaptive degradation parameter is expressed as
H' is an adaptive degradation factor, h is the same as a traditional non-local mean value, the value of the parameter h is related to the noise variance of the image, and is an empirical value, generally, h is taken as gamma sigma, wherein the value range of gamma is [90,100 ]]Better effect can be achieved. Since the variation coefficient of the homogeneous region is close to the noise variance and the variation coefficient of the heterogeneous region is much larger than the noise variance, the parameter τ is selected from a value calculated when the position of the symmetric point of the sigmoid curve is equal to the noise variance, that is, when CV is σ, τ + ξ (CV/CV) is equal to σmax) Solving for tau-xi (CV/CV) by 0max) The parameter xi is used for controlling the inclination of the S-shaped curve, and the range of the obtained parameter xi is [20,50 ] through a large number of experiments]It is noted that this step is calculated for the global variation coefficient of the entire image, where the parameter represents the global maximum variation coefficient of the image. CV ofmaxThe maximum local variation coefficient of the whole image is used for normalizing the variation coefficient.
And 5, constructing a self-adaptive filtering window based on the variation coefficient.
In this embodiment, step 5 is implemented by the following process:
according to the step 2, it can be known that speckle noise of the SAR image can be simplified into multiplicative noise, and if the noise is white gaussian noise with a mean value of unity, according to the multiplicative noise model of the step B, the variation coefficient of one image block can be represented as
Where Ψ is an image block in the observation image v, and E (u) if the image block is in a homogeneous regionΨ 2)＝[E(uΨ)]2The coefficient of variation can be simplified to
Since the noise mean value is unity, it can be obtained
That is, therefore, the image block variation coefficients in the homogeneous region can be approximated as the image noise variance; if the image block is in a heterogeneous area, E (u)Ψ 2)＞＞[E(uΨ)]2Therefore, the image block variation coefficient in the heterogeneous region is far larger than the image noise variance, that is, the value of the variation coefficient at the edge texture is far larger than that of the homogeneous region, and by utilizing the characteristic of the local variation coefficient of the image, a self-adaptive anisotropic two-dimensional filtering window can be constructed, the anisotropic filtering window is introduced into the distance measurement, so that the precision of the similarity measurement is increased, the pixel points with high similarity can be selected more effectively, and the edge texture target and the point target can be protected more effectively; meanwhile, in the homogeneous region, because the variation coefficient is approximate to the noise variance, and the filtering window is approximate to flat, isotropic filtering can be ensured, namely, under the condition that the homogeneous region is not influenced, the edge texture direction information is fully utilized for weighting, and the direction details of the image are effectively captured. The anisotropic filtering window constructed from the image local variation coefficients can be expressed as:
ACV(i,k)＝exp(-|CV(i,k)-CV(i)|2)
wherein A isCVAnd (i, k) is an anisotropic weighting window taking i as a center, and CV (i, k) is a variation coefficient of the kth pixel point in the image block taking i as a center. If i is located on the texture, the edge pixel can be assigned with high weight value by using the variation coefficient, and the pixel point irrelevant to the edge direction is assigned with low weight value, thereby achieving the purpose of direction selection, then ACV(i, k) forming a filtering window along the image texture; if i is located on the point target, the variation coefficient of the point target is far larger than that of the surrounding pixel points, so that a central weighted form can be obtained, and ACV(i, k) the distribution resembles a gaussian filter window; if i is in the homogeneous region, since the variation coefficients of the homogeneous region are close to the noise variance, each pixel point is almost endowed with the same weight, and A isCV(i, k) is a flat filter window with uniform distribution.
And 6, obtaining the value of the pixel point based on the non-local algorithm of the heterogeneity measure.
Referring to fig. 2, step 6 is implemented by the following process:
step 61, calculating image block similarity measure
According to the characteristic that coherent speckles can be modeled into multiplicative noise in a simplified mode, the similarity measurement is calculated by using the ratio measurement, and in combination with the self-adaptive filtering window constructed in the step 5, for image blocks taking i and j as centers in a search area, the similarity measurement is expressed as
Wherein d isCV(i, j) is the similarity measure of the image blocks with the pixel points i and j as the center, Ψ is the image block,
the function xi (x) is a normalization factor of the coefficients of variation in the image block, and is expressed by xi (x) x +1/x, r (i, j) ui(k)/uj(k) The gray scale of the kth pixel point of the image block taking i and j as the centerValue ui(k) And uj(k) The ratio between r (i, j) and r (i, j) is shown to be closer to 1, indicating that the two gray values are closer together, and further away from 1, indicating that the two gray values differ more, and that the function xi () is a function [0, + ∞ ]]The function is decremented and then incremented and the minimum is taken when r (i, j) equals 1.
Step 62, calculating the weight of the pixel point
According to the NLM algorithm, a Gaussian kernel function is used for calculating the weight of the pixel points, the similarity measure of the image blocks is used as a distance, the self-adaptive degradation coefficient calculated in the step 4 is used as the degradation coefficient, normalization is carried out, and the weight among the pixel points can be expressed as
Wherein,
for the normalization coefficient, h' is an adaptive degradation factor for controlling the decay rate of the exponential function.
Step 63, calculating pixel values of pixel points
According to the NLM algorithm, for a pixel point i, a search area omega is formediThe pixel points in the image are overlapped according to the weight value to obtain the estimated value of the pixel point i
Wherein u (j) is the gray value corresponding to the pixel j.
The invention estimates the pixel value of each pixel point as NLCV[u](i) In that respect The method can effectively keep the texture edge information while filtering out the coherent spots.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.
Claims (5)
1. A non-local mean filtering method for a synthetic aperture radar image is characterized by comprising the following steps:
step 1, obtaining a synthetic aperture radar amplitude image and initializing parameters, wherein the initialized parameters comprise a search window size Ds, an image window size Ds, a degradation parameter coefficient gamma and an S curve gradient parameter xi;
step 2, calculating to obtain speckle distribution of the image, and obtaining noise variance of the speckle based on the speckle distribution;
step 3, calculating an image variation coefficient according to the standard deviation and the mean value of the image;
step 4, calculating self-adaptive degradation parameters according to the image variation coefficient; the method comprises the following steps:
adaptively adjusting the coefficient of variation based on an S-curve represented as
Wherein the parameter tau is used for controlling the offset of the S curve, and the parameter xi is used for controlling the inclination of the S curve;
according to the variation coefficient, for each pixel point in the image, the self-adaptive degradation parameter is expressed as
Wherein h' is an adaptive degradation factor, h ═ γ σ, CVmaxThe maximum local variation coefficient of the whole image;
step 5, constructing a self-adaptive filtering window based on the variation coefficient; the method comprises the following steps:
according to the multiplicative noise model in the step 2, the noise is white Gaussian noise with the mean value as a unit value, and the variation coefficient of one image block is expressed as
Where Ψ is an image block in the observation image v, and E (u) if the image block is in a homogeneous regionΨ 2)＝[E(uΨ)]2And the coefficient of variation is simplified to
Based on the noise value as a unit value
The image block variation coefficient in the homogeneous region is the image noise variance, and the anisotropic filtering window constructed based on the local variation coefficient of the image is expressed as
ACV(i,k)＝exp(-|CV(i,k)-CV(i)|2)
Wherein A isCV(i, k) is an anisotropic weighting window taking i as a center, and CV (i, k) is a variation coefficient of a kth pixel point in an image block taking i as a center;
and 6, obtaining the value of the pixel point based on the similarity measure of the image block.
2. The non-local mean filtering method for synthetic aperture radar images according to claim 1, wherein the step 1 further comprises the following steps:
setting a pixel point i as a pixel positioned at an image position i, setting an image block i as an image block with the size ds and the pixel point i as the center, and searching a region omegaiThe image blocks are image blocks with the size Ds and with the pixel point i as the center.
3. The non-local mean filtering method for synthetic aperture radar images according to claim 2, wherein the step 2 comprises the following steps:
when the image pixel spacing and the radar imaging resolution unit are the same, the noise power is uncorrelated, and the coherent speckle is incoherent multiplicative noise expressed as
υ＝u·n
Wherein upsilon is an observed value, u is a radar Reflection Coefficient (RCS), n is speckle noise, and the equivalent visual number of the image can be obtained according to the standard deviation and the mean value of the image
Wherein, L is equivalent vision, sigma is standard deviation of the image, mu is mean value of the image, the coherent speckles are completely developed, and the distribution function of the intensity condition is
Wherein Gamma (·) is Gamma function, radar reflection coefficient is set to 1, and completely developed coherent speckle noise distribution is obtained
The variance of the noise intensity is calculated as
The amplitude value of the image is the square root of the intensity, and the amplitude condition distribution of the obtained observation value is
Setting the radar reflection coefficient to 1, the distribution function of amplitude noise can be obtained
The noise variance is calculated as
4. The non-local mean filtering method for synthetic aperture radar images according to claim 3, wherein the step 3 comprises the following steps:
coefficient of variation of
Wherein, CV is a local variation coefficient of the pixel point.
5. The non-local mean filtering method for synthetic aperture radar images according to claim 1, wherein the step 6 comprises the following steps:
step 61, calculating image block similarity measure
Based on a ratio measure and the adaptive filtering window, a similarity measure is expressed as
Wherein d isCV(i, j) is the similarity measure of the image blocks with the pixel points i and j as the center, psi is the image block, ZCV(i) The function xi (x) is a normalization factor of the coefficients of variation in the image block, and has an expression xi (x) x +1/x, ui(k) The k-th pixel point of the image block taking i as the centerGray value of uj(k) The gray value of the kth pixel point of the image block taking j as the center;
step 62, calculating the weight of the pixel point
According to the NLM algorithm, based on the Gaussian kernel function, calculating the weight between the pixel points
Wherein,
is a normalized coefficient;
step 63, calculating pixel values of pixel points
According to the NLM algorithm, for a pixel point i, a search area omega is formediThe pixel points in the image are overlapped according to the weight value to obtain the estimated value of the pixel point i
Wherein u (j) is the gray value corresponding to the pixel j.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810502002.9A CN109035152B (en) | 2018-05-23 | 2018-05-23 | Non-local mean filtering method for synthetic aperture radar image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810502002.9A CN109035152B (en) | 2018-05-23 | 2018-05-23 | Non-local mean filtering method for synthetic aperture radar image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109035152A CN109035152A (en) | 2018-12-18 |
| CN109035152B true CN109035152B (en) | 2022-03-18 |
Family
ID=64611660
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810502002.9A Active CN109035152B (en) | 2018-05-23 | 2018-05-23 | Non-local mean filtering method for synthetic aperture radar image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109035152B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490815A (en) * | 2019-07-12 | 2019-11-22 | 西安理工大学 | A kind of region adaptivity denoising method based on division Bregman algorithm |
| CN110570379B (en) * | 2019-09-11 | 2023-03-24 | 重庆大学 | Non-local mean value CT image noise reduction method based on structure tensor |
| CN110852975B (en) * | 2019-11-15 | 2023-10-17 | 大连海事大学 | Method for driving non-local mean value of radar image by combining variation coefficient and local entropy |
| 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 |
| CN111447426B (en) * | 2020-05-13 | 2021-12-31 | 中测新图(北京)遥感技术有限责任公司 | Image color correction method and device |
| CN111783583B (en) * | 2020-06-22 | 2024-02-20 | 西安电子科技大学 | SAR image speckle suppression method based on non-local mean algorithm |
| CN112098997B (en) * | 2020-09-18 | 2021-10-15 | 欧必翼太赫兹科技(北京)有限公司 | Three-dimensional holographic imaging security inspection radar image foreign matter detection method |
| CN112381727B (en) * | 2020-11-03 | 2023-12-15 | 中航航空电子有限公司 | Image denoising method and device, computer equipment and storage medium |
| CN112750077B (en) * | 2020-12-14 | 2023-01-20 | 中国船舶重工集团公司第七0九研究所 | Parallelized synthetic aperture radar image sub-band splicing effect processing method |
| CN113160084B (en) * | 2021-04-19 | 2022-07-01 | 新疆大学 | Denoising method and device for quantum dot fluorescence image on surface of porous silicon biosensor |
| CN113160213B (en) * | 2021-05-12 | 2022-10-14 | 西南科技大学 | Novel pixel membership estimation method fusing local and non-local information of image |
| CN113759375B (en) * | 2021-09-15 | 2024-02-09 | 云南师范大学 | SAR image non-local mean denoising method based on statistical characteristics |
| CN115035326B (en) * | 2022-06-09 | 2024-04-12 | 电子科技大学 | Radar image and optical image accurate matching method |
| CN116703920B (en) * | 2023-08-07 | 2023-10-20 | 济宁市华祥石墨制品有限公司 | Graphite thermal insulation barrel crack detection method based on image processing |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521811A (en) * | 2011-12-09 | 2012-06-27 | 中国人民解放军海军航空工程学院 | Method for reducing speckles of SAR (synthetic aperture radar) images based on anisotropic diffusion and mutual information homogeneity measuring degrees |
| CN103077499A (en) * | 2013-01-09 | 2013-05-01 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image pre-processing method based on similar block |
| CN103886563A (en) * | 2014-04-18 | 2014-06-25 | 中南民族大学 | SAR image speckle noise inhibition method based on non-local mean and heterogeneity measurement |
| CN103983973A (en) * | 2014-05-28 | 2014-08-13 | 西安电子科技大学 | Synthetic aperture radar imaging method based on image sparse domain noise distribution constraint |
| CN104730500A (en) * | 2015-02-04 | 2015-06-24 | 电子科技大学 | Synthetic aperture radar residual range migration correction method |
| CN105184749A (en) * | 2015-09-18 | 2015-12-23 | 中南民族大学 | Heterogeneity pre-correction-based wavelet domain SAR image despeckling method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9262808B2 (en) * | 2013-02-07 | 2016-02-16 | Mitsubishi Electric Research Laboratories, Inc. | Denoising of images with nonstationary noise |
-
2018
- 2018-05-23 CN CN201810502002.9A patent/CN109035152B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521811A (en) * | 2011-12-09 | 2012-06-27 | 中国人民解放军海军航空工程学院 | Method for reducing speckles of SAR (synthetic aperture radar) images based on anisotropic diffusion and mutual information homogeneity measuring degrees |
| CN103077499A (en) * | 2013-01-09 | 2013-05-01 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image pre-processing method based on similar block |
| CN103886563A (en) * | 2014-04-18 | 2014-06-25 | 中南民族大学 | SAR image speckle noise inhibition method based on non-local mean and heterogeneity measurement |
| CN103983973A (en) * | 2014-05-28 | 2014-08-13 | 西安电子科技大学 | Synthetic aperture radar imaging method based on image sparse domain noise distribution constraint |
| CN104730500A (en) * | 2015-02-04 | 2015-06-24 | 电子科技大学 | Synthetic aperture radar residual range migration correction method |
| CN105184749A (en) * | 2015-09-18 | 2015-12-23 | 中南民族大学 | Heterogeneity pre-correction-based wavelet domain SAR image despeckling method |
Non-Patent Citations (6)
| Title |
|---|
| A non-local algorithm for image denoising;Antoni Buades等;《2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR"05)》;20050725;1-6 * |
| Feature-Based Nonlocal Polarimetric SAR Filtering;Xiaoli Xing等;《remote sensing》;20171013;1-9 * |
| Learning a Dilated Residual Network for SAR Image Despeckling;Qiang Zhang等;《arXiv:1709.02898》;20170909;1-18 * |
| 基于自适应 Bilateral滤波的SAR图像相干斑抑制;李光廷等;《电子与信息学报》;20120515;第34卷(第5期);第1076-1081页 * |
| 基于自适应非局部均值的SAR图像相干斑抑制;陈世媛等;《系统工程与电子技术》;20170926;第39卷(第12期);第2683-2690页 * |
| 基于非局部均值滤波的SAR图像去噪;易子麟等;《电子与信息学报》;20120415;第34卷(第4期);第950-955页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109035152A (en) | 2018-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109035152B (en) | Non-local mean filtering method for synthetic aperture radar image | |
| CN109919870B (en) | SAR image speckle suppression method based on BM3D | |
| CN107301661A (en) | High-resolution remote sensing image method for registering based on edge point feature | |
| CN106338733B (en) | Forward-Looking Sonar method for tracking target based on frogeye visual characteristic | |
| Yuan et al. | Noise reduction for sonar images by statistical analysis and fields of experts | |
| CN110837768A (en) | Rare animal protection oriented online detection and identification method | |
| KR101891410B1 (en) | Apparatus for classifying target of inverse synthetic aperture radar image using trace transform and method thereof | |
| CN102156971B (en) | Speckle suppression method of synthetic aperture radar (SAR) image based on linear singularity information | |
| CN113362293A (en) | SAR image ship target rapid detection method based on significance | |
| CN103020922A (en) | PCA (principal component analysis) transformation based SAR (synthetic aperture radar) image speckle suppression method | |
| CN102999908A (en) | Synthetic aperture radar (SAR) airport segmentation method based on improved visual attention model | |
| CN107831473B (en) | Distance-instantaneous Doppler image sequence noise reduction method based on Gaussian process regression | |
| CN103871031A (en) | Kernel regression-based SAR image coherent speckle restraining method | |
| CN112836707B (en) | ISAR image aerial target length feature extraction method | |
| Vitale et al. | Edge preserving CNN SAR despeckling algorithm | |
| Devapal et al. | Discontinuity adaptive SAR image despeckling using curvelet-based BM3D technique | |
| CN112669332A (en) | Method for judging sea and sky conditions and detecting infrared target based on bidirectional local maximum and peak local singularity | |
| CN112488960A (en) | SAR image speckle suppression method based on boundary and context constraint | |
| CN109285148B (en) | Infrared weak and small target detection method based on heavily weighted low rank and enhanced sparsity | |
| CN114155425B (en) | Weak and small target detection method based on Gaussian Markov random field motion direction estimation | |
| CN112734666B (en) | SAR image speckle non-local mean suppression method based on similarity value | |
| Kondaveeti et al. | Abridged shape matrix representation for the recognition of aircraft targets from 2D ISAR imagery | |
| CN114429593A (en) | Infrared small target detection method based on rapid guided filtering and application thereof | |
| James et al. | Blind estimation of single look side scan sonar image from the observation model | |
| Chen et al. | Texture Enhancement Method of Oceanic Internal Waves in SAR Images Based on Non-local Mean Filtering and Multi-scale Retinex |
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 |