CN119168913B

CN119168913B - Defocused speckle deblurring method based on Wiener filtering

Info

Publication number: CN119168913B
Application number: CN202411231600.9A
Authority: CN
Inventors: 董博; 黄海昇; 倪梓浩; 白玉磊; 谢胜利
Original assignee: Guangdong University of Technology
Current assignee: Guangdong University of Technology
Priority date: 2024-09-04
Filing date: 2024-09-04
Publication date: 2025-09-23
Anticipated expiration: 2044-09-04
Also published as: CN119168913A

Abstract

The application provides a defocusing speckle deblurring method based on wiener filtering, which comprises the steps of collecting an original speckle image, including a reference image and at least 1 deformation image, evaluating speckle quality in the original speckle image through an average intensity gradient to obtain an evaluation result, deblurring the deformation image with lower speckle quality according to the evaluation result to obtain a corresponding restored image of the deformation image, carrying out virtual grid division on the reference image to obtain at least 2 subregions, acquiring correlation between any subregion and the deformation image to obtain an optimal matching position, obtaining full-field displacement corresponding to any deformation image according to the optimal matching position, further obtaining full-field strain, and obtaining more strain results in a larger depth direction by comparing the strain results before deblurring and after deblurring, thereby realizing range expansion related to digital images.

Description

Defocusing speckle deblurring method based on wiener filtering

Technical Field

The application belongs to the field of non-contact material mechanics detection, and particularly relates to a defocusing speckle deblurring method based on wiener filtering.

Background

The related technology of the digital image is one of the non-contact optical measurement methods with application prospect at present, can provide in-plane displacement analysis and strain calculation by tracking the deformation of speckle patterns on the surface of a material, has the advantages of no damage, high precision, high timeliness and the like, and is widely applied to the fields of material science, mechanical engineering, civil engineering and the like. However, when the image correlation analysis is performed by using the technique, the imaging depth range is affected, and when the displacement amount of the material in the depth direction exceeds the range of the optical system, the speckle image of the material is blurred due to defocusing. Such ambiguity can significantly increase the difficulty of feature matching between subsequent images, and may even lead to matching failure, thereby limiting the range of accurate measurement in the depth direction. In the fields of structural dynamic testing, material fatigue analysis, structural health monitoring, etc., a wide range of depth measurement capabilities are helpful in revealing the behavior of materials and structures under extreme conditions.

To overcome this limitation and increase the range of measurement, improvements may be made to the imaging system. For example, the depth scale may be extended by increasing the aperture of the lens or using a multi-focal fusion technique. Large aperture lenses can increase the depth of field range to capture clear images over a larger depth range, but doing so requires specially designed lenses and increases system volume and cost, and multi-focus fusion techniques by capturing multiple images at different focus positions and combining them into one high depth of field image, but this approach typically requires additional hardware support, such as a multi-camera system, resulting in a significant increase in cost.

Therefore, a defocusing and speckle deblurring method capable of realizing speckle tracking of a digital image correlation technology under a defocusing condition and overcoming the problem of speckle image correlation mismatch caused by limited range is needed to be researched and developed.

Disclosure of Invention

The application aims to provide a defocusing speckle deblurring method based on wiener filtering, which is used for at least solving one technical problem in the prior art.

The technical scheme of the application is as follows:

An out-of-focus speckle deblurring method based on wiener filtering, comprising:

Collecting an original speckle image, wherein the original speckle image comprises a reference image and at least 1 deformed image, and evaluating the speckle quality in the original speckle image through an average intensity gradient to obtain an evaluation result;

deblurring the deformed image with lower speckle quality according to the evaluation result to obtain a corresponding restored image of the deformed image;

performing virtual grid division on the reference image to obtain at least 2 sub-areas, and acquiring the correlation between any sub-area and the deformed image to obtain the optimal matching position;

and by comparing the strain results before and after deblurring, the method can obtain more strain results in a larger depth direction, thereby realizing the range expansion related to the digital image.

And deblurring the deformed image with lower speckle quality according to the evaluation result to obtain a corresponding restored image of the deformed image, wherein the method comprises the following steps of:

initializing a blur kernel radius and a signal-to-noise ratio corresponding to speckles in the original speckle image according to the evaluation result:

Acquiring an initial restored image according to the blur kernel radius and the signal-to-noise ratio;

setting a standard for restoring the fuzzy speckle image, and adjusting the values of the fuzzy kernel radius and the signal-to-noise ratio to obtain a corresponding restored image of the deformed image.

The obtaining an initial restored image according to the blur kernel radius and the signal-to-noise ratio comprises the following steps:

assuming that the blur kernel radius is h, the blur model is expressed as g=h×f+n;

wherein g is a blurred image, f is an original speckle image, x represents convolution operation, and n is noise;

the estimated value of the original speckle image is expressed as:

Wherein F' is an estimated value of an original speckle image, H is Fourier transform of a blur kernel H, H ² is an amplitude spectrum of the blur kernel, SNR is a signal-to-noise ratio, and G represents Fourier transform of a blur image G.

The setting of the standard of the restoration of the fuzzy speckle image, and the adjustment of the values of the fuzzy kernel radius and the signal-to-noise ratio comprise the following steps:

taking the half-width of the gray level histogram of the clear speckle image as a standard for recovering the blurred speckle image, thereby adjusting the values of the blur kernel radius and the signal-to-noise ratio, wherein the half-width FWHM of the gray level histogram is expressed as:

FWHM=D_right-D_left;

Wherein D _right and D _left are respectively gray values corresponding to when the first one of the right and left sides in the gray histogram reaches or exceeds half the peak height.

Said adjusting values of said blur kernel radius and signal to noise ratio comprising:

Gradually increasing the blur kernel radius R and reducing the signal-to-noise ratio SNR to obtain the half-width of the gray level histogram of the restored image;

Comparing the half-width of the gray level histogram of the current restored image with the half-width of the clear speckle image in the focal length to obtain a comparison result;

and continuously adjusting the values of the blur kernel radius R and the signal-to-noise ratio SNR according to the comparison result until the half-width of the gray level histogram of the restored image reaches the half-width of the gray level histogram of the clear speckle image in the focal length, and taking the blur kernel radius R and the signal-to-noise ratio SNR at the moment as optimal values.

When adjusting the values of the blur kernel radius and the signal-to-noise ratio, the method further comprises:

and carrying out weighted average processing on any pixel in the restored image by combining the spatial proximity and the gray level similarity weight so as to adjust the blur kernel radius and the ringing effect caused by the signal-to-noise ratio to obtain a final restored image, wherein the method comprises the following steps:

determining a spatial proximity weight by acquiring a spatial distance between any pixel in the restored image and a pixel in a neighborhood thereof;

determining gray scale similarity weight by acquiring gray scale value similarity of any pixel in the restored image and pixels in a neighborhood of the pixel;

And for the pixel, using the spatial proximity weight and the gray level similarity weight to acquire a weighted average value of all pixels in the neighborhood, and taking the weighted average value as a final value of the pixel, so as to inhibit ringing effect and obtain a final restored image.

The virtual meshing of the reference image is performed to obtain at least 2 sub-areas, and the correlation between any sub-area and the deformed image is obtained to obtain the best matching position, which comprises the following steps:

Selecting an interested region from the reference image as a search window, and performing virtual grid division on the interested region to obtain at least 2 sub-regions;

taking the subareas formed by virtual grids as rigid motions, and acquiring the correlation between any subarea and the deformed image;

And searching sub-pixel levels of the speckle images by using a bilinear interpolation method to obtain gray values corresponding to image functions of the deformed images at coordinates (x ', y') until the best matching position is obtained.

The method for searching the sub-pixel level of the speckle image by using the bilinear interpolation method to obtain the gray value corresponding to the image function of the deformed image at the coordinates (x ', y') comprises the following steps:

Let f (x ₁,y₁)、f(x₁,y₂)、f(x₂,y₂)、f(x₂,y₁) denote the gray values corresponding to the image functions of the deformed image at coordinates (x ₁,y₁)、(x₁,y₂)、(x₂,y₂)、f(x₂,y₁), respectively;

obtaining linear interpolation in the x' direction:

Obtaining linear interpolation in the y' direction:

And then the gray value g (x ', y') is obtained.

The obtaining full-field displacement corresponding to any deformed image according to the optimal matching position, obtaining full-field strain, includes:

Taking a sub-region center point Q (x ₀,y₀) in the reference image as a center, and searching in the deformed image to obtain the best matching position Q ₁(x₀+u,y₀ +v);

obtaining full-field displacement according to the change of any point P1 (x, y) in the sub-region of the deformed image, wherein the full-field displacement comprises the following steps:

The change of the deformed image at any point P1 (x, y) in the sub-region is represented by P (x, y) in the reference image sub-region using a zero-order displacement characterization function: wherein u and v are displacements in the x direction and the y direction respectively;

Estimating a displacement ladder by using a center difference method to obtain full field strain:

Wherein, the AndThe rate of change of the displacement u in the x-direction and the y-direction are indicated respectively,AndThe change rate of v in the x and y directions is shown, u (x, y) and v (x, y) are shown, the displacement in the x and y directions is shown at point (x, y), Δx and Δy are steps, ε _xx and ε _yy are positive strains in the x and y directions, respectively, and ε _xy is a shear strain.

The evaluating the speckle quality in the original speckle image by the average intensity gradient comprises:

Acquiring an average gray gradient delta of the speckle image:

where W denotes a speckle pixel width, H denotes a speckle pixel height, f _x (x, y) denotes a first derivative of the pixel point (x, y) in the x direction, and f _y (x, y) denotes a first derivative of the pixel point (x, y) in the y direction.

The beneficial effects of the application at least comprise:

The method comprises the steps of firstly collecting an original speckle image, including a reference image and at least 1 deformation image, evaluating speckle quality in the original speckle image through an average intensity gradient to obtain an evaluation result, then deblurring the deformation image with lower speckle quality according to the evaluation result to obtain a corresponding restored image of the deformation image, then carrying out virtual grid division on the reference image to obtain at least 2 subareas, acquiring correlation between any subarea and the deformation image to obtain an optimal matching position, and finally obtaining full-field displacement corresponding to any deformation image according to the optimal matching position to obtain full-field strain so as to improve the digital image range of the original speckle image. According to the method, the original image is restored by carrying out self-adaptive adjustment on the filtering parameters of the fuzzy type and the intensity of the speckle, so that speckle tracking of the digital image correlation technology under the defocus condition can be stably realized, and the problem of speckle image correlation mismatch caused by limited range is solved.

Drawings

FIG. 1 is a flow chart of the method of the present application;

FIG. 2 is a gray level histogram;

FIG. 3 is a schematic diagram of a bilinear interpolation algorithm;

fig. 4 is a graph of digital image dependent strain calculations.

Detailed Description

The application is further described below with reference to the accompanying drawings.

Specific example I:

The present application provides an embodiment:

referring to FIG. 1, a defocusing speckle deblurring method based on wiener filtering specifically comprises the following steps:

Firstly cutting a white tested sample into a required shape and size, polishing the surface of the sample, then cleaning the surface of an object by using clear water and an organic solvent to enhance the adhesive force of the surface of the sample to speckle, and finally using matt black paint to spray the speckle to form a speckle pattern with black and white contrast on the surface of the object. S2, reference image acquisition:

Before a specific load is applied to a sample, the matched image acquisition software is utilized to acquire a speckle pattern with high contrast and definition, the speckle pattern is used as an initial condition of the sample in a natural state, and 1 reference image is acquired.

S3, load is applied:

Under the condition that the field of view of the camera is unchanged, the two-dimensional linear displacement platform is used for controlling the sample to move at fixed intervals, so that the camera obtains a deformed speckle image, and an original speckle image is obtained.

S4, acquiring a deformed image:

In the sample loading process, a camera and matched image acquisition software are used for acquiring the surface of the sample for a plurality of times under the condition of fixed acquisition frame rate, so as to obtain a series of deformed images.

S5, speckle quality assessment:

After a series of deformed images are acquired, these images are evaluated for quality. The average gray gradient delta of the speckle image is calculated as follows:

Where W denotes a speckle pixel width, H denotes a speckle pixel height, f _x (x, y) denotes a first derivative of the pixel point (x, y) in the x direction, and f _y (x, y) denotes a first derivative of the pixel point (x, y) in the y direction. The larger the average intensity gradient, the better the speckle image quality. And if the result does not reach the target value, executing step 7 to deblur the speckle image, and then continuing the subsequent calculation.

S6, speckle deblurring:

The method comprises the following specific steps:

S601, initializing a blur kernel radius R and a signal-to-noise ratio SNR:

Since the original clear image corresponding to the blurred image is unknown, the size and the signal-to-noise ratio of the blur kernel cannot be directly obtained. In this case, a reasonable initial value needs to be set first. To facilitate the optimization of subsequent R and SNR, the present patent selects a smaller R and a larger SNR as initial values.

S602, wiener filtering to deblur:

On the basis of the initialized blur kernel radius R and the signal-to-noise ratio SNR, a blur model is firstly established, and according to the characteristics of an imaging system, defocusing blur exists in a speckle image outside the depth of field. For defocus blur, the blur kernel is typically represented as a gaussian function. Assuming the blur kernel is h, the blur model can be expressed as:

g=h*f+n (2)

Where g is a blurred image, f is an original image, x represents a convolution operation, and n is noise. Fourier transforming it can result in the expression of the fuzzy model in the frequency domain as:

G=H·F+N (3)

wherein G, H, F, N represent the fourier transforms of g, h, f, n, respectively. The purpose of deblurring is to obtain an estimate of the original image, which can be expressed in the frequency domain as:

F'=H_w·G (4)

where F' is an estimated value of the original image, and H _w is a wiener filtered transfer function, which can be simplified as:

Where H (f) is the Fourier transform of the blur kernel H, |H (f) ² | is the magnitude spectrum of the blur kernel and SNR is the signal-to-noise ratio. Thus, the estimate of the original image can be expressed as:

Substituting the blur kernel radius R and the signal-to-noise ratio SNR in the step 1 into the formula (5) to obtain an initial restored image F ₁.

S603, optimizing the blur kernel radius R and the signal-to-noise ratio SNR:

In order to obtain a restored image that is more clear and accurate than the original restored image F ₁, optimization of R and SNR is required. Throughout the defocus process, the blurred speckle image has a smaller gray-level histogram half-width than the sharp speckle image. And the half-width of the restored gray level histogram of the blurred speckle image obtained at the defocusing position is close to the half-width of the gray level histogram of the clear speckle image obtained in the focal distance. Therefore, the gray-level histogram half-width of the clear speckle image obtained in the focal length is used as a standard for restoration of the blurred speckle image, so that the proper R and SNR are selected. The full width at half maximum FWHM of the gray-scale histogram is shown in fig. 2, which can be expressed as:

FWHM=D_right-D_left (7)

The specific optimization process is that ① gradually increases the blur kernel radius R and reduces the signal-to-noise ratio SNR, the step (6-II) is repeated after each adjustment, and the half-width of the gray level histogram of the restored image is calculated. ② And comparing the half-width after each adjustment with the half-width of the clear speckle image in the focal length, and selecting R and SNR which enable the half-width to be close to the clear image as optimal values to obtain a restored image F _n.

S604. an adaptive local smoothing strategy to suppress ringing effects:

The above steps can generate obvious ringing phenomenon in the restored image F _n, which is specifically represented by a stripe or band structure with alternate brightness and darkness near the edge of the speckle image, and the phenomenon can cause detail distortion of the image and affect the image quality. This is because wiener filtering involves deconvolution of the blur kernel of the image. If the blurring kernel contains zero or near zero frequency components, deconvolution may result in amplification of the high frequency components, thereby producing a ringing effect. The spectral characteristics of an image determine its composition at different spatial frequencies. Ringing may also occur during recovery if the original image has a strong abrupt change at high frequencies, where an adaptive local smoothing strategy is employed to suppress the ringing. The adaptive local smoothing strategy effectively suppresses ringing by combining spatial proximity and gray scale similarity weights and performing a weighted average process on each pixel.

The specific process is as follows:

s6041. spatial proximity weight:

The spatial proximity weight is determined by calculating the spatial distance of each pixel from the pixels in its neighborhood. The closer a pixel is to the center pixel, the greater its weight. The spatial proximity weight w _s (q, p) can be expressed as:

where p represents the center pixel, q represents a pixel in the field, σ _d is the standard deviation of the spatial proximity weight. S6042, gray level similarity weight:

The gray scale similarity weight is determined by calculating the gray scale value similarity of each pixel to the pixels in its neighborhood. The closer the gray value of the pixel is to the center pixel, the greater its weight. The gray scale similarity weight w _r (q, p) can be expressed as:

Wherein I (p) and I (q) are the gray values of the center pixel and the pixels in the neighborhood, respectively, σ _r is the standard deviation of gray-scale similarity weights.

S6043. weighted average:

For each pixel, a weighted average of all pixels in the neighborhood is calculated using the spatial proximity and gray scale similarity weights described above as the final value for that pixel. The pixel value I' (p) after the ringing suppression process can be expressed as:

where N (p) represents the neighborhood of pixel p. After suppressing ringing, the final restored image F is obtained.

S7, searching and positioning the image:

and carrying out digital image correlation analysis on the restored image F. Firstly, selecting a region of interest in a reference image as a search window, and performing virtual meshing on the region of interest, wherein all sub-regions formed by virtual meshing are regarded as rigid motion. And then, calculating the correlation between the deformed image and each reference subarea by using a standardized covariance cross-correlation function, wherein the closer the calculation result is to 1, the more accurate the matching result is represented. The normalized covariance cross-correlation function can be expressed as:

Wherein f (x, y) and g (x ', y') represent gray values corresponding to image functions of the reference image and the deformed image at coordinates (x, y) and (x ', y'), respectively. The search at the sub-pixel level of the speckle image is then performed using bilinear interpolation to yield g (x ', y'), as shown in fig. 3.

Let f (x ₁,y₁)、f(x₁,y₂)、f(x₂,y₂)、f(x₂,y₁) denote the gray values corresponding to the image functions of the deformed image at coordinates (x ₁,y₁)、(x₁,y₂)、(x₂,y₂)、f(x₂,y₁), respectively, and the calculation process is as follows:

s701, calculating linear interpolation in the x' direction:

S702, calculating linear interpolation in the y' direction:

Substituting the formula (12) and the formula (13) into the formula (14) can obtain the gray value g (x ', y').

S8, reconstructing displacement and strain fields:

And searching to obtain the best matching position Q ₁(x₀+u,y₀ +v by taking the center point Q (x ₀,y₀) of the reference image subarea as the center through the last step, wherein u and v are displacements in the x direction and the y direction respectively. The change in any point P ₁(x^*,y^* in this sub-region of the deformed image can be represented by P (x, y) in the reference image sub-region using a zero-order displacement characterization function:

After calculation for each sub-region, full field displacement can be obtained. The displacement ladder is then estimated using the center difference method to obtain full field strain, calculated as follows:

Wherein, the AndThe rate of change of the displacement u in the x-direction and the y-direction are indicated respectively,AndThe change rate of v in the x and y directions is shown, u (x, y) and v (x, y) are shown, the displacement in the x and y directions is shown at point (x, y), Δx and Δy are steps, ε _xx and ε _yy are positive strains in the x and y directions, respectively, and ε _xy is a shear strain. Application example:

The sample is controlled to move from 0mm to 6.0mm by a two-dimensional displacement platform, and speckle images are acquired every 0.5mm of movement. When the displacement exceeds 4.5mm, the speckle image starts to appear blurred, and the speckle quality is poor. In order to verify the feasibility of the proposed method. Speckle images with depth scales of 4.5mm to 6.0mm were analyzed. Fig. 4 shows the results of strain calculations in the x-direction for the digital image correlation technique before and after deblurring the speckle image, and compares the errors in both cases. Fig. 4 (a) shows the result of strain calculation associated with a digital image without performing the speckle image deblurring process. It can be seen that as the depth-direction displacement amount increases, the degree of blurring of the acquired speckle image becomes larger and larger. When the displacement amount in the depth direction reaches 6.0mm, the speckle image matching fails due to the large degree of blurring of the speckle image, and the strain cannot be calculated, as shown in fig. 4 (a-4). And after deblurring the speckle image using the methods mentioned herein, the quality of the speckle image is improved, and the matching can be successful at the location where the original speckle image failed, as shown in fig. 4 (b-4). Next, error calculation is performed on the strain calculation results in both cases, as shown in fig. 4 (c). It can be seen that after deblurring treatment by the method of the patent, the error of the strain calculation result is reduced, and when the depth displacement reaches 6.0mm, a better strain calculation result can be obtained, so that the expansion of the range of the digital image correlation in the depth direction is realized.

Specific example II:

The application also provides an embodiment:

An electronic device comprising a storage medium and a processing unit, wherein the storage medium is for storing a computer program, the processing unit is in data exchange with the storage medium, and the processing unit is used for executing the computer program by the processing unit when performing defocusing speckle deblurring, so as to perform the steps of the defocusing speckle deblurring method based on wiener filtering as described in the specific embodiment I.

Specific example III:

a computer readable storage medium having a computer program stored therein, which when executed, performs the steps of the wiener filtering-based defocus speckle deblurring method of embodiment I.

In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present application, however, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, with the computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

The foregoing disclosure is merely illustrative of some embodiments of the application, and the application is not limited thereto, as modifications may be made by those skilled in the art without departing from the scope of the application. The above-mentioned inventive sequence numbers are merely for description and do not represent advantages or disadvantages of the implementation scenario.

Claims

1. An out-of-focus speckle deblurring method based on wiener filtering, comprising:

Obtaining full-field displacement corresponding to any deformed image according to the optimal matching position, further obtaining full-field strain, and improving the range related to the digital image for the original speckle image by comparing the strain results before and after deblurring;

setting a standard for restoring the fuzzy speckle image, and adjusting the values of the fuzzy kernel radius and the signal-to-noise ratio to obtain a corresponding restored image of the deformed image;

Establishing a fuzzy model, wherein defocusing blur exists in a speckle image outside a depth of field range, the fuzzy core is expressed as a Gaussian function for the defocusing blur, and the fuzzy model is expressed as follows under the assumption that the radius of the fuzzy core is h:

g=h*f+n;

the estimated value of the original speckle image is expressed as:

wherein F' is an estimated value of an original speckle image, H is Fourier transform of a fuzzy core H, H ² is an amplitude spectrum of the fuzzy core, and SNR is a signal-to-noise ratio;

FWHM=D_right-D_left;

Wherein, D _right and D _left are respectively gray values corresponding to the first one of the right side and the left side in the gray histogram reaching or exceeding half of the peak value;

Acquiring an average gray gradient delta of the speckle image:

2. The wiener filtering-based out-of-focus speckle deblurring method of claim 1, wherein said adjusting values of the blur kernel radius and signal-to-noise ratio comprises:

3. The wiener filtering-based out-of-focus speckle deblurring method of claim 1, further comprising, when adjusting the values of the blur kernel radius and signal-to-noise ratio:

4. The wiener filtering-based out-of-focus speckle deblurring method of claim 1, wherein said virtually meshing said reference image to obtain at least 2 sub-regions, and obtaining a correlation of any one of said sub-regions with said deformed image to obtain an optimal matching position, comprises:

And searching the sub-pixel level of the speckle image by using a bilinear interpolation method to obtain the gray value corresponding to the image function of the deformed image at the coordinates (x ', y') until the best matching position is obtained.

5. The wiener filtering-based out-of-focus speckle deblurring method of claim 4, wherein said searching at a speckle image subpixel level using bilinear interpolation results in a gray value for an image function of said distorted image at coordinates (x ', y'), comprising:

obtaining linear interpolation in the x' direction:

Obtaining linear interpolation in the y' direction:

And then the gray value g (x ', y') is obtained.

6. The wiener filtering-based defocusing speckle deblurring method of claim 1, wherein said obtaining full-field displacement corresponding to any of said deformed images according to said best matching position, further obtaining full-field strain, comprises: