A Speckle Image Matching Algorithm Based on the Light Intensity Change Model

Abstract

The present invention discloses a speckle image matching algorithm based on an intensity change model. S100: Collect speckle images before and after deformation with speckle features on the surface; S200: Select a calculation region and set basic parameters for image matching; S300: Establish an intensity change model and use linear least squares estimation (IM-LLSE) to obtain the integer-pixel matching position; S400: Use the intensity coefficient and the integer-pixel position as initial values, and combine the intensity change model to perform sub-pixel matching using the inverse composite Gauss-Newton method (IM-ICGN); S500: Transmit the matching parameters of the nodes that have completed matching to the neighboring sub-regions, and finally complete the sub-pixel matching of all nodes within the entire calculation region. The method of the present invention is suitable for application in scenarios with bright and dark changes in the environment for deformation tracking and shape reconstruction of the object surface. Compared with the current mainstream algorithms, it can improve the matching accuracy and reliability while having comparable computational efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of experimental mechanics, and particularly to a speckle image matching algorithm based on a light intensity change model. Background Art

[0002] The digital image correlation method has been widely used in the measurement of the deformation and strain of an object surface under force and three-dimensional topography scanning. In a mechanics experiment, black and white random speckles are sprayed on the surface to be measured, and the full-field deformation before and after the force can be obtained through a matching algorithm; in topography scanning, random speckles are projected on the surface to be measured, and through the matching algorithm and three-dimensional reconstruction, it has the advantage of high-dynamic fast scanning.

[0003] Technically, the core of the digital image correlation method is the matching algorithm, that is, for the given speckle images before and after deformation or the left and right cameras in binocular vision, the matching of corresponding points on the two images is realized based on the algorithm. The process of finding corresponding points on the two speckle images mainly adopts the ICGN algorithm based on the ZNSSD correlation function at present. However, in this algorithm, the meaning of the ZNSSD function as a correlation function is not clear, and it is considered that the gray mapping relationship between the reference sub-region and the deformed sub-region is linear, which is not conducive to being applied to the occasions with complex changes before and after illumination. In view of this, the present invention introduces a light intensity change model as the correlation function and derives the calculation methods for whole-pixel matching and sub-pixel matching. Summary of the Invention

[0004] The purpose of the present invention is to provide a speckle image matching algorithm based on a light intensity change model. By introducing a light intensity change model into the mainstream algorithm, the physical meaning of each link in the correlation matching is made more clear, and the relevant model can be defined targeted, so that the matching is more accurate. The method of the present invention is suitable for the deformation tracking and topography reconstruction of the object surface under the scene of changing light and darkness in the environment. Compared with the current mainstream algorithms, it can improve the matching accuracy and reliability while having the same calculation efficiency.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A speckle image matching algorithm based on a light intensity change model is specifically implemented according to the following steps:

[0006] S100. Collect the speckle images before and after the deformation with speckle features on the surface to be measured;

[0007] S200. Select a calculation area and set the basic parameters for image matching. The set basic parameters include sub-region size, node step size, and seed points;

[0008] S300. For the seed points, perform a whole-pixel search for similar sub-regions on the deformed image; establish an intensity change model, and use linear least squares estimation I M-LLSE to obtain the whole-pixel matching position. The intensity change model is also called the gray-scale mapping model, and one of the linear or quadratic parabola models is selected; on the deformed image, determine the whole-pixel position corresponding to the deformed node according to the minimum regression deviation.

[0009] S400. For the seed points, perform a sub-pixel search for similar sub-regions on the deformed image. Take the intensity coefficient and the whole-pixel position as the initial values, combine with the intensity change model, and use the inverse composite Gauss-Newton method I M-I CGN for sub-pixel matching; during the sub-pixel matching process, use the idea of variable parameter iteration. First, fix the intensity coefficient, iterate to find the deformation coefficient, and then calculate the new intensity coefficient according to the deformation coefficient, and repeat until the iteration converges.

[0010] S500. Transmit the matching parameters of the already matched nodes to the neighboring sub-regions, and finally complete the sub-pixel matching of all nodes within the entire calculation region; then the coordinate difference between the corresponding nodes before and after deformation is the deformation amount at that node.

[0011] Preferably, in step S300, the intensity change model is introduced into the similarity function, and the form is Correspondingly, it is briefly recorded as f i = Φ(g i ). Any model of the intensity change model can be selected according to the actual situation.

[0012] Preferably, the whole-pixel search method in step S300 is specifically as follows:

[0013] S301. The intensity linear change model is f i = a × g i + b; establish a similarity function C pssd = ∑(a × g i + b - f i ) 2 ;

[0014] S302. In the search region on the deformed image, perform the correlation calculation of the reference sub-region f i and the deformed sub-region g' i point by point. Let ∑(a × g' i + b - f i ) 2 = 0. The values of a and b can be obtained through linear least squares estimation, and then the residual R = ∑(a × g' i + b - f i ) 2 ;

[0015] S303. Compare the residuals corresponding to all nodes within the search area. The sub-region corresponding to the minimum residual is the matching sub-region for integer pixels. At the same time, the corresponding light intensity coefficients a and b are relatively accurate gray-scale mapping coefficients for the reference sub-region and the integer-pixel deformed sub-region.

[0016] Preferably, in step S400, the sub-pixel search algorithm IM-ICGN, that is, the specific steps of variable parameter iteration are as follows:

[0017] S401. Take the integer-pixel node coordinates and light intensity coefficients obtained in S303 as the initial values for Gauss-Newton iteration, and set the initial values of other deformation parameters to 0.

[0018] S402. For the relevant function after adding the shape function, the light intensity linear change model is: f i = a × g i + b, and the expression is where a and b are light intensity change coefficients, f(x + W(ξ; Δp)) is the gray-scale description of the sub-region before deformation, g(x + W(ξ; p)) is the gray-scale description of the sub-region after deformation; use the Gauss-Newton method IC-GN to iterate once to obtain the deformation parameters and the corresponding deformed sub-region, and then use the gray-scale of the deformed sub-region and the reference sub-region to find the new and accurate coefficients a and b through the linear least squares estimation in S302;

[0019] S403. Take the new deformation parameters and light intensity coefficients obtained by iteration in step S402 as the initial values, and use the Gauss-Newton method for iterative calculation again to obtain new deformation parameters and light intensity coefficients. Repeat this process, and perform convergence judgment at the end of each iteration. If the accuracy requirement is met, the sub-pixel matching ends.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] A speckle image matching algorithm based on a light intensity change model proposed by the present invention makes the physical meaning of each link in the relevant matching clearer by introducing a light intensity change model into the mainstream algorithm, and can thus define the relevant model specifically to make the matching more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 are the steps of the matching algorithm of the present invention;

[0023] Figure 2 is the point-to-point gray-scale mapping model inside the reference sub-region and the deformed sub-region of the present invention;

[0024] Figure 3 is the integer-pixel search process based on linear intensity change of the present invention;

[0025] Figure 4Steps of the sub-pixel iterative algorithm based on the intensity model of the present invention;

[0026] Figure 5 The reference image and its calculation region in the embodiment of the present invention;

[0027] Figure 6 The calculation result on the deformed image in the embodiment of the present invention;

[0028] Figure 7 The deformation field in the horizontal direction on the deformed image in the embodiment of the present invention;

[0029] Figure 8 The deformation field in the vertical direction on the deformed image in the embodiment of the present invention. Detailed implementation manners

[0030] Next, the technical solutions of the present invention in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] Embodiment

[0032] A speckle image matching algorithm based on the light intensity change model of the present invention first gives the calculation steps of the algorithm during single-point matching, and then shows the application of this method in actual measurement experiments:

[0033] (1) Calculation steps

[0034] The core of the present invention is Figure 1 the part shown by the dashed box in, including the integer-pixel search algorithm (IM-LLSE) and the sub-pixel variable parameter iterative algorithm (IM-ICGN). The basis of these two parts is to establish a gray-scale mapping model between the reference sub-region and the deformed sub-region. As Figure 2 shown, the present invention believes that the point-to-point mapping model inside the reference sub-region f i and the deformed sub-region g i includes intensity mapping and deformation mapping. The intensity mapping can be described in the following format:

[0035]

[0036] Briefly recorded as f i = Φ(g i ), such as: f i = q1 × g i + q0, f i = q1ln(1 + gi ) + q0, etc.

[0037] Wherein:

[0038] Reference sub-region center x = (x0, y0, 1) T , the offset of the internal position of the sub-region relative to the center ξ = (Δx, Δy, 1) T ,

[0039] Deformation parameter p = (u, u x , u y , v, v x , v y ) T , deformation correction parameter Δp = (Δu, Δu x , Δu y , Δv, Δv x , Δv y ) T ,

[0040] Deformation mapping

[0041] Mapping correction

[0042] Theoretically, the light intensity change model can be constructed by the sum of several single-valued functions, that is q j is the light intensity change coefficient.

[0043] The following describes the formulas for integer-pixel and sub-pixel matching. Let the half side length of the reference sub-region be m pixels, and the sub-region contains N = (2m + 1) 2 pixels.

[0044] (1.1) Integer-pixel search algorithm IM-LLSE

[0045] The essence of the integer-pixel search algorithm in the present invention is linear least squares estimation. After obtaining the estimation coefficients, substitute them back into the model and compare the minimum value of the fitting residual. The corresponding deformed sub-region is the integer-pixel matching sub-region. Figure 3 The integer-pixel search process based on linear intensity change is shown as follows. The following specifically describes the calculation method.

[0046] (a) If the light intensity change model is set as f i = q1g i + q0, denote q = (q1, q0) T , then the correlation function is Wherein, when calculating the similarity between the reference sub-region and the candidate deformed sub-region point by point in the search area of the deformed image, construct the equation system Briefly denoted as gq = f.

[0047] (b) If the light intensity change model is set as Let \(q=(q_2,q_1,q_0)\) T , then the corresponding correlation function is Construct the system of equations Briefly denoted as \(g_q = f\).

[0048] The least squares solution of the above system of equations can be expressed as \(q=(g\) T \(g)\) -1 (g T \(f)\). Define the fitting deviation Calculate the fitting deviations of all candidate deformed sub-regions. The deformed sub-region corresponding to the smallest fitting deviation is the deformed sub-region after integer pixel matching. Its light intensity coefficient and the central offset \((u, v)\) relative to the reference sub-region can be used as the initial values for sub-pixel matching iteration.

[0049] (1.2) Sub-pixel matching algorithm IM-ICGN

[0050] Since the present invention adds a light intensity change model on the basis of the traditional method, the corresponding ICGN algorithm is improved. The overall idea is to perform linear least squares estimation after each IC-GN iteration to update the value of the light intensity coefficient as the initial value for the next iteration. The idea of this two-step method can be seen specifically in Figure 4 as shown. The following specifically describes the calculation method.

[0051] Based on the intensity change model \(f\) i \(=\varPhi(g\) i ) construct the objective expression:

[0052]

[0053] Briefly denoted \(H\) remains unchanged during iteration. Then the iteration format is:

[0054]

[0055] where the first-order gray gradient

[0056] (briefly denoted as )

[0057] The updated form of the shape function is:

[0058]

[0059] Iteration convergence condition: \(\vert\vert W(\xi;p\) k+1 ) - W(\xi;p\) k )\(\vert\vert\leq0.001\) and \(\vert\vert q\) (k+1) \(-q\)(k) ||≤0.001

[0060] After each iteration, the light intensity parameters of the reference subset f(x+ξ) and the updated deformed subset g(x+W(ξ; p)) are estimated using the method in integer pixel matching (linear least squares estimation).

[0061] (a) If the linear intensity variation function f is used i =q1g i +q0 constructs the target expression:

[0062]

[0063] The iteration format is:

[0064]

[0065] (b) If the quadratic parabola light intensity variation model is used Construct the target expression:

[0066]

[0067] The iteration format is:

[0068]

[0069] (2) Implementation of measurement

[0070] Conventional mechanical measurement experiments typically use a camera to capture speckle images of the surface being measured before and after deformation, and then perform matching calculations. Here, a sand deposition experiment is used as an example to illustrate the application of the present invention. A complete measurement experiment includes the following steps:

[0071] First, build a measurement environment based on the size of the measured area and the parameters of the camera and lens. For details on the estimation of the measurement distance and the selection of the lens, please refer to the patent "A visual measurement method for full-field motion tracking of rotating objects" Figure 3 .

[0072] Secondly, spots are made on the surface of the object to be measured; since the sand in this experiment has natural texture, it can be measured directly.

[0073] Next, the speckle image of the deformation process is collected. Figure 5 and Figure 6 The background photos are images of sand collected before and after deformation.

[0074] Finally, the deformation field matching calculation is performed. According to the calculation method of the present invention, the calculation area is delineated on the reference image, and the sub-area and step size are set to obtain Figure 5The crosses marked in [text] each represent the center (node) of a sub-region. Figure 5 The central sub-region of the calculation region in [text] is set as the seed sub-region. After the matching is completed, the calculation parameters are passed to the surrounding sub-regions. The corresponding nodes obtained after the calculation are as Figure 6 the cross marks in [text]. The coordinate differences of the paired nodes are solved to obtain the deformation fields in the horizontal and vertical directions as Figure 7 , Figure 8 shown.

[0075] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A speckle image matching algorithm based on an optical intensity change model, characterized in that, The implementation is specifically carried out according to the following steps: S100. Collect the speckle images before and after deformation of the measured surface with speckle features; S200. Select the calculation area and set the basic parameters for image matching. The set basic parameters include sub-region size, node step size, and seed points; S300. For the seed points, perform a full-pixel search for similar sub-regions on the deformed image; establish an intensity change model, and use linear least squares estimation IM-LLSE to obtain the full-pixel matching position. The intensity change model is also called the gray mapping model, and one of the linear or quadratic parabola models is selected; on the deformed image, determine the full-pixel position corresponding to the deformation node according to the minimum regression deviation; S400. For the seed points, perform a sub-pixel search for similar sub-regions on the deformed image. Take the intensity coefficient and full-pixel position as the initial values, combine with the intensity change model, and use the inverse combined Gaussian-Newton method IM-ICGN for sub-pixel matching; during the sub-pixel matching process, use the idea of variable parameter iteration. First, fix the intensity coefficient, iterate to find the deformation coefficient, and then find the new intensity coefficient according to the deformation coefficient, and repeat until the iteration converges; S500. Transfer the matching parameters of the nodes that have been matched to the neighboring sub-regions, and finally complete the sub-pixel matching of all nodes within the entire calculation area; then the coordinate difference between the corresponding nodes before and after deformation is the deformation amount at that node; The step S300 introduces the light intensity change model into the similarity function in the form of where x is the center of the reference sub-region, ξ is the offset of the internal position of the sub-region relative to the center, and is correspondingly abbreviated as f i = Φ(g i ), and any model of the light intensity change model can be selected according to the actual situation; The specific full-pixel search method in step S300 is as follows: S301. The light intensity linear change model is f i = a × g i + b; Establish a similarity function C including the light intensity coefficient pssd = ∑(a × g i + b - f i ) 2 ; S302. In the search area on the deformed image, perform point-by-point calculation on the reference sub-region f i and the deformed sub-region g i to make ∑(a×g i +b - f i ) 2 = 0. The values of a and b can be obtained through linear least squares estimation, and then the residual R = ∑(a×g i +b - f i ) 2 ; S303. Compare the residuals corresponding to all nodes in the search area. The sub-region corresponding to the minimum residual is the full-pixel matching sub-region; at the same time, the corresponding intensity coefficients a and b are the relatively accurate gray mapping coefficients of the reference sub-region and the full-pixel deformed sub-region; The specific steps of the sub-pixel search algorithm IM-ICGN in step S400, that is, variable parameter iteration, are as follows: S401. Take the full-pixel node coordinates and intensity coefficients obtained in S303 as the initial values of the Gaussian-Newton iteration, and set the initial values of other deformation parameters to 0; S402. For the relevant functions after adding the shape function, the light intensity linear change model is: f i = a × g i + b, and the expression is where a and b are light intensity change coefficients, f(x + W(ξ; Δp)) is the gray-scale description of the sub-region before deformation, g(x + W(ξ; p)) is the gray-scale description of the sub-region after deformation; the deformation parameters and the corresponding deformed sub-regions are obtained by using the Gauss-Newton method IC-GN for a single iteration, and then the new and accurate coefficients a and b are obtained by linear least squares estimation in S302 using the gray scales of the deformed sub-region and the reference sub-region; S403. Take the new deformation parameters and intensity coefficients obtained by iterating in step S402 as the initial values, and use the Gaussian-Newton method for iterative calculation again to obtain new deformation parameters and intensity coefficients. Repeat this process, and perform a convergence judgment at the end of each iteration. If the accuracy requirement is met, the sub-pixel matching ends.

Images