An edge-preserving feature InSAR image denoising method based on adaptive window

Abstract

The non-local mean filtering method is a SAR interferogram denoising processing means capable of keeping image edge features, and the non-local mean filtering method widely used at present generally uses a window with fixed size and shape to filter the whole image, which can cause mutual influence between targets with different scattering characteristics and destroy the detail information of the interferogram. In view of the problem, the application provides an edge feature preserving InSAR image denoising method based on an adaptive window, the method can match an adaptive direction window according to the characteristics of the SAR interferogram, keep the texture information of the interferogram to be filtered, and adaptively adjust the size of the filtering window according to the amplitude information of the SAR interferogram, so as to reduce the mutual influence between targets with different scattering characteristics, keep more detail information on the premise of meeting the filtering requirement, and thus realize high-quality SAR interferogram filtering.

Description

Technical Field

[0001] This invention belongs to the field of synthetic aperture radar technology, and particularly relates to an edge-preserving InSAR image denoising method based on adaptive windows. Background Technology

[0002] Synthetic Aperture Radar (SAR) is a high-resolution imaging radar with advantages such as all-weather, all-time, and wide-area observation capabilities, and is widely used in the field of surface observation. The basic principle of Synthetic Aperture Radar Interferometric (InSAR) imaging technology is to generate a SAR interferogram using two SAR images with a certain viewing angle difference, and extract the elevation information of the observed target based on the interferometric phase.

[0003] Due to their imaging principles, SAR images are often contaminated by speckle noise, which degrades the quality of SAR images, thereby affecting the quality of SAR interferograms and reducing the accuracy of target observation. To achieve high-precision target observation, filtering SAR interferograms is a common noise reduction technique.

[0004] Nonlocal mean filtering is a filtering method that preserves image edge features. The main idea behind this algorithm is to leverage the high redundancy and decorrelation of noise in natural images. In a natural image, any edge, point, or other feature can be found in multiple similar parts within the same image. The nonlocal mean filtering process involves traversing all pixels in the entire image for each pixel to be filtered, determining the similarity, assigning weights based on the similarity, and then performing a weighted average to obtain the filtered pixel. The advantage of this algorithm is its ability to better preserve the detailed features of the image.

[0005] Currently widely used nonlocal mean filtering methods typically use a window of fixed size and shape to filter the entire image. However, for SAR interferograms, the scattering characteristics of observed targets in the scene vary considerably. Using a fixed filtering window can cause targets with different scattering characteristics to interfere with each other, destroying the detailed information of the interferogram. Most proposed improved nonlocal mean filtering methods do not address improvements to the filtering window itself. Therefore, when processing SAR interferograms, the size and shape of the filtering window cannot be adjusted according to the amplitude information and texture features of the interferogram, resulting in poor filtering performance. Summary of the Invention

[0006] To address the aforementioned SAR interferogram filtering problem, this invention proposes an edge-preserving InSAR image denoising method based on an adaptive window. This method can match a suitable orientation window according to the inherent characteristics of the SAR interferogram and adaptively adjust the size of the filtering window based on the amplitude information of the SAR interferogram, thereby achieving high-quality SAR interferogram filtering.

[0007] The input to this design method is a SAR interferogram S and a signal-to-noise ratio (SNR) threshold. th The filtering parameter h, and 16 directional window templates [W1, W2, ..., W 16 The output is the filtered SAR interferogram complex matrix.

[0008] The method for SAR interferogram filtering of the present invention includes:

[0009] Step 1: Input the SAR interferogram S to be filtered and calculate the signal-to-noise ratio of the entire image.

[0010] Input the SAR interferogram S to be filtered, and select a flat region with low amplitude as the noise reference region S. N According to the amplitude of the complex image S and S N The average amplitude of the region is used to calculate the signal-to-noise ratio (SNR) of each pixel in the entire image.

[0011] Step 2: Select pixels to be filtered based on their signal-to-noise ratio.

[0012] Pixels to be filtered are selected based on the calculated signal-to-noise ratio (SNR) of the pixels and the input SNR threshold, where SNR(i,j) ≥ SNR. th The pixels do not need to be filtered.

[0013] Step 3: Calculate the correlation coefficient of the pixel to be filtered based on the signal-to-noise ratio.

[0014] The correlation coefficient of the pixel to be filtered is calculated based on the signal-to-noise ratio, using the following formula:

[0015]

[0016] Step 4: Simulate the trend of the standard deviation of the interferometric phase as a function of the number of interferogram views under the signal-to-noise ratio condition of the point to be filtered.

[0017] Based on the correlation coefficient and probability density function of the interference phase obtained in the previous step, the variation trend of the standard deviation of the interference phase with the number of interferogram views under the corresponding signal-to-noise ratio conditions is simulated.

[0018] The probability density function of the interference phase is:

[0019]

[0020] Where L is the number of views in the interferogram, and the function Γ is defined as follows:

[0021]

[0022] 2F1(·) is a hypergeometric function, and the definition of a hypergeometric function is as follows:

[0023]

[0024] The formula for calculating the standard deviation of the interference phase is as follows:

[0025]

[0026] Where, μ φ =φ0, which is the mean value of the interference phase.

[0027] Step 5: Record the number of interferogram views when the standard deviation of the interferometric phase reaches the set threshold.

[0028] With SNR = SNR th The phase standard deviation is used as a threshold, and the number of views L that makes the interference phase standard deviation reach the threshold is recorded under the signal-to-noise ratio (SNR) (i,j) of the pixel to be filtered, x = S(i,j). th .

[0029] Step 6: Match and select the appropriate orientation window

[0030] A rectangular window with a side length of 9 is constructed with the pixel to be filtered x as the center point, based on 16 input directional window templates [W1, W2, ..., W...]. 16 Calculate the mean; the direction window template diagram is provided by... Figure 2 As shown, only pixels in the white areas of the window participate in the calculation. The window that maximizes the average amplitude of the pixels within the window is selected as the direction window W for the pixel x to be filtered. x .

[0031] Step 7: Calculate similarity weights

[0032] The similarity weight between each pixel in the directional window and the pixel to be filtered is calculated using the following formula.

[0033]

[0034] Here, v(x) is a rectangular image patch centered at the pixel x to be filtered, with a side length of 3; v(y) is another rectangular image patch centered at another pixel y in the direction window selected in the previous step, also with a side length of 3. This is the normalization coefficient.

[0035] Step 8: Filter the appropriate number of pixels and perform a weighted average.

[0036] Filter the top L values ​​with the highest similarity weight from the direction window. th The weighted average of each pixel is used to filter a single pixel. The formula for the weighted average is as follows:

[0037]

[0038] Step 9: Traverse all pixels in the image to be filtered to complete the filtering process.

[0039] Repeat the above steps to filter all the pixels in the image.

[0040] This completes all the steps.

[0041] The beneficial effects of this invention are as follows:

[0042] This invention enables the matching of appropriate directional window templates based on the characteristics of the original SAR interferogram, and adaptively determines the size of the filtering window based on the amplitude information of the interferogram. This allows the filtered SAR interferogram to retain more detailed information while meeting the filtering requirements, and reduces the mutual influence between targets with different scattering characteristics, thereby improving the filtering quality and enhancing the accuracy of target observation. Attached Figure Description

[0043] Figure 1 This is a flowchart of the steps of the present invention;

[0044] Figure 2 This is a schematic diagram of the direction window template input in this invention;

[0045] Figure 3 This is an unfiltered SAR interferometric phase map;

[0046] Figure 4 The image shows a SAR interferometric phase map obtained after filtering using the method described in this invention.

[0047] Figure 5 This represents the unfiltered local interference amplitude;

[0048] Figure 6 The local interference amplitude is obtained after filtering using the method in this invention. Detailed Implementation

[0049] The present invention will now be described in detail.

[0050] In this invention, the user input is a SAR interferogram S to be filtered, and the signal-to-noise ratio threshold SNR is... th The filtering parameter h, and 16 directional window templates [W1, W2, ..., W 16 The output is a filtered SAR interferogram. The specific steps are as follows:

[0051] Step 1: Input the SAR interferogram S to be filtered and calculate the signal-to-noise ratio of the entire image.

[0052] Input the SAR interferogram S to be filtered, and select a flat region with low amplitude as the noise reference region S in the imaging area of ​​the entire interferogram. N , will S N The average interference amplitude within the range is used as a noise energy reference value, and the signal-to-noise ratio of the entire interferogram is calculated pixel by pixel.

[0053] Step 2: Select pixels to be filtered based on their signal-to-noise ratio.

[0054] In SAR interferograms, pixels with high signal-to-noise ratios (SNR) correspond to very low noise levels and are often targets with strong scattering characteristics. For such targets, good observation results can be obtained without filtering. Therefore, filtering conditions are set based on SNR; when the pixel SNR meets the following conditions, filtering is not required.

[0055] SNR(i,j)≥SNR th (8)

[0056] Step 3: Calculate the correlation coefficient of the pixel to be filtered based on the signal-to-noise ratio.

[0057] To achieve the function of adaptively improving the window using amplitude information from SAR interferograms, it is necessary to correlate the signal-to-noise ratio (SNR) of the SAR interferogram with the correlation coefficient. The formula for calculating the correlation coefficient based on the SNR is as follows:

[0058]

[0059] Step 4: Simulate the trend of the standard deviation of the interferometric phase as a function of the number of interferogram views under the signal-to-noise ratio condition of the point to be filtered.

[0060] Based on the signal-to-noise ratio (SNR) of the point to be filtered, the trend of phase standard deviation with the number of views is simulated according to the correlation coefficient and probability density function of the interference phase calculated in the previous step. Under different SNR conditions, the phase standard deviation decreases with the increase of the number of views. The probability density function of the interference phase is:

[0061]

[0062] Where L is the number of views in the interferogram, and the function Γ is defined as follows:

[0063]

[0064] 2F1(·) is a hypergeometric function, and the definition of a hypergeometric function is as follows:

[0065]

[0066] The formula for calculating the standard deviation of the interference phase is as follows:

[0067]

[0068] Where, μ φ =φ0, which is the mean value of the interference phase.

[0069] Step 5: Record the number of interferogram views when the standard deviation of the interferometric phase reaches the set threshold.

[0070] According to the filtering requirements, the phase standard deviation of the filtered points must reach the standard where filtering is not required. Therefore, SNR = SNR th The phase standard deviation at time is used as a threshold, and the number of views L that makes the interference phase standard deviation reach the threshold is recorded under the signal-to-noise ratio condition of the pixels to be filtered. th .

[0071] The multi-look process is the process of averaging the range or azimuth of a SAR image. Therefore, the number of views in a multi-look process can be considered as the number of pixels participating in the averaging. This is similar to the size of the filter window in nonlocal mean filtering. In nonlocal mean filtering, the size of the filter window is the number of pixels that ultimately participate in the weighted average. Therefore, the number of views can be considered equivalent to the size of the search window in nonlocal mean filtering.

[0072] Step 6: Match and select the appropriate orientation window

[0073] Interference fringes in SAR interferometric phases exhibit strong directionality, and rectangular filtering windows can easily disrupt the continuity of fringes in dense fringe regions. Therefore, using a directional window that matches the direction of the interference fringes is more suitable for filtering SAR interferograms. The pixel values ​​within the directional window are relatively close and can be approximated as homogeneous. Filtering with a directional window can remove noise without disrupting the interference fringes.

[0074] This method uses 16 different directional window templates, which are based on a rectangular window with a side length of 9, and are divided into different directional structures. When selecting a directional window, the mean value of the 16 input directional window templates needs to be calculated. Only the white pixels in the window are included in the calculation. Finally, the window that maximizes the mean value of the pixels within the window is selected as the directional window for the pixels to be filtered.

[0075] Step 7: Calculate similarity weights

[0076]

[0077] Here, v(x) is a rectangular image patch centered at the pixel x to be filtered, with a side length of 3; v(y) is another rectangular image patch centered at another pixel y in the direction window selected in the previous step, also with a side length of 3. This is the normalization coefficient.

[0078] Step 8: Filter the appropriate number of pixels and perform a weighted average.

[0079] Filter the top L values ​​with the highest similarity weight from the direction window. th The weighted average of each pixel is used to filter a single pixel. The weighted average formula is as follows:

[0080]

[0081] Step 9: Traverse all pixels in the image to be filtered to complete the filtering process.

[0082] The following provides an implementation example with specific parameters.

[0083] In this example, the SAR interferometric image to be filtered is generated from two preprocessed (registration, deskewing, phase compensation) TerraSAR-X images. The imaging area is the Regent Beijing Hotel and some surrounding buildings. The unfiltered SAR interferometric phase image is... Figure 3 Based on experience, the filter parameter is set to h=2, and the signal-to-noise ratio threshold is set to SNR. th =15dB, the input direction window template is as follows Figure 2 .

[0084] First, in step 1, the interferogram to be filtered is input and the overall signal-to-noise ratio is calculated. Then, in step 2, the pixels to be filtered are selected. Next, steps 3, 4, and 5 determine the size of the filtering window. In step 6, a suitable directional window is matched and selected. Finally, steps 7, 8, and 9 perform SAR interferogram filtering to obtain the filtered interferogram.

[0085] Figure 4 The SAR interferometric phase image after filtering by the method in this invention is shown. It can be seen that the noise level in the filtered interferometric phase is reduced, while the detailed information and texture structure of the original SAR interferometric phase are preserved, and the filtering effect is good.

[0086] Furthermore, to verify the edge-preserving effect of this method, the local interference amplitudes before and after filtering were extracted and compared, as shown below. Figure 5 and Figure 6 In the filtering results obtained using the method of this invention, the edges and details of the original interferogram are almost completely preserved, effectively retaining the texture details.

[0087] Of course, there are many other embodiments of the present invention. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A method for edge-preserving InSAR image denoising based on adaptive windows, characterized in that, Includes the following steps: Step 1: Input the SAR interferogram to be filtered And calculate the signal-to-noise ratio of the entire image; Step 2: Select the pixels to be filtered based on their signal-to-noise ratio; Step 3: Calculate the correlation coefficient of the pixels to be filtered based on the signal-to-noise ratio. The formula is: ； Step 4: Simulate the trend of the standard deviation of the interferometric phase as a function of the number of interferogram views under the signal-to-noise ratio condition of the point to be filtered; Based on the signal-to-noise ratio of the point to be filtered, the variation trend of the phase standard deviation with the number of views is simulated according to the correlation coefficient and the probability density function of the interference phase calculated in the previous step; the probability density function of the interference phase is: ； in, It is the number of views of the interferogram, a function. The definition is as follows: ； These are hypergeometric functions, defined as follows: ； The formula for calculating the standard deviation of the interference phase is as follows: ； in, , is the mean of the interference phase; Step 5: Record the number of interferogram views when the standard deviation of the interferometric phase reaches the set threshold. Record the number of views when the standard deviation of the interferometric phase reaches the threshold under the signal-to-noise ratio conditions of the pixels to be filtered. ; Step 6: Match and select the appropriate orientation window; Step 7: Calculate the similarity weights; the formula for calculating the similarity weights between each pixel in the direction window and the pixel to be filtered is: ； in, Therefore, the pixels to be filtered A rectangular image patch centered on the image, with a side length of 3; It is another pixel in the direction window selected in the previous step. Another rectangular image patch centered on the center, also with a side length of 3. , is the normalization coefficient; Step 8: Filter a corresponding number of pixels and perform a weighted average; select the pixels with the highest similarity weight from the directional window. The weighted average of each pixel is used to filter a single pixel. The formula for the weighted average is as follows: ； Step 9: Traverse all pixels in the image to be filtered to complete the filtering process.

2. The edge-preserving InSAR image denoising method based on adaptive windowing as described in claim 1, characterized in that, In step 2, the pixels to be filtered are selected based on the calculated signal-to-noise ratio (SNR) of the pixels and the input SNR threshold. The pixels do not need to be filtered.

3. The edge-preserving InSAR image denoising method based on adaptive windowing as described in claim 1, characterized in that, In step 5, The phase standard deviation at time is used as a threshold and recorded in the pixel to be filtered. signal-to-noise ratio The number of eyes that makes the standard deviation of the interference phase reach the threshold under the condition .

Images