An imaging self-calibration method, system, electronic device and storage medium based on multi-scale displacement constraint

Abstract

This invention relates to the field of image processing and discloses an imaging self-calibration method, system, electronic device, and storage medium based on multi-scale displacement constraints. The method includes: acquiring a beam image sequence; constructing a logarithmic domain difference model to transform the beam image; constructing a dimensionality-reduced multi-scale global matrix using effective overlapping pixel pairs and reconstructing the image to obtain the true relative beam intensity distribution; and using the true relative beam intensity distribution to correct the beam image, obtaining a calibrated image. This invention completely eliminates the systematic errors introduced by calculation using displacement constraints, achieving high precision and artifact-free results. The smoothness of the calibrated image is significantly better than traditional methods. Furthermore, it eliminates the need to purchase film or water tanks, fully utilizing existing radiotherapy equipment. The process can be automatically executed by the treatment bed under software control, requiring no consumables and being fully automated, greatly reducing equipment calibration costs and time. It is applicable to all two-dimensional radiotherapy detectors based on integral modes, demonstrating strong equipment versatility and universality.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically, to a self-calibration method, system, electronic device, and storage medium for imaging devices based on multi-scale displacement constraints. Background Technology

[0002] The pixel response consistency (i.e., "flat field characteristics") of two-dimensional radiotherapy imaging equipment is fundamental to ensuring the accuracy of dose verification. Currently available imaging calibration methods include the external flat field source method and the physical reference source method.

[0003] One method, the external flat-field source method, involves illuminating the detector with an accelerator whose output beam is almost flat, thereby calculating the detector's gain coefficient. However, in reality, it is impossible to find an accelerator that is absolutely flat, which introduces the accelerator's own non-uniformity error into the flat-field correction results. Therefore, this method suffers from insufficient accuracy.

[0004] The physical reference source method uses radiochromic film or three-dimensional water tank scan data as a standard reference field to calculate the detector's gain coefficient. This method suffers from drawbacks such as expensive consumables, cumbersome procedures, and high error rates. For example, film is costly and its processing is complex, requiring scanning and calibration, with scanning often taking up to 24 hours. Furthermore, water tank setup is time-consuming, typically requiring several hours, which cannot meet the needs of frequent, rapid calibration. The data obtained from the film or water tank needs to be registered and aligned with the imaging results, a process highly susceptible to human error. Summary of the Invention

[0005] To address the aforementioned shortcomings, this invention proposes a fully automated, high-precision imaging self-calibration method that does not require an external reference source. By introducing constraint displacement to construct a fully connected constraint network, and combining it with a multi-scale global matrix with split-axis dimensionality reduction, a high-precision beam distribution and detector gain matrix are obtained, thereby achieving flat-field correction of the electronic imaging device.

[0006] In a first aspect, the present invention discloses an imaging self-calibration method based on multi-scale displacement constraints, comprising the following steps: Acquire a reference image and perform image acquisition based on the sampling displacement of at least two constraints to obtain a beam image sequence containing the reference image and the displacement image; A logarithmic difference model is constructed, and the beam image sequence is transformed using the logarithmic difference model to obtain the logarithmic distribution of beam intensity difference corresponding to pixel displacement; Based on the logarithmic distribution of beam intensity difference and effective overlapping pixel pairs, a dimension-reduced multi-scale global matrix is ​​constructed, and the dimension-reduced multi-scale global matrix is ​​reconstructed and restored to obtain the true relative beam intensity distribution. The gain matrix is ​​obtained by using the real relative beam intensity distribution and the reference image, and the beam image to be calibrated is then calibrated to obtain the calibration image.

[0007] Secondly, the present invention also provides an imaging self-calibration system based on multi-scale displacement constraints, comprising: The data acquisition module is used to acquire a reference image and perform image acquisition based on the sampling displacement of at least two constraints to obtain a beam image sequence containing the reference image and the displacement image; The model building module is used to construct a logarithmic domain difference model and use the logarithmic domain difference model to transform the beam image sequence to obtain the logarithmic distribution of the beam intensity difference corresponding to the pixel displacement. The split-axis dimensionality reduction and reconstruction module is used to construct a dimensionality-reduced multi-scale global matrix based on the logarithmic distribution of beam intensity difference and effective overlapping pixel pairs, and to reconstruct and restore the dimensionality-reduced multi-scale global matrix to obtain the true beam intensity distribution. The self-calibration module obtains the gain matrix using the real relative beam intensity distribution and reference image, and then calibrates the beam image to be calibrated to obtain the calibration image.

[0008] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the imaging self-calibration method based on multi-scale displacement constraints described above.

[0009] Fourthly, the present invention also provides a non-transient computer-readable storage medium storing computer instructions for causing a computer to implement the imaging self-calibration method based on multi-scale displacement constraints described above.

[0010] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are as follows: The imaging self-calibration scheme provided by this invention completely eliminates systematic errors introduced by calculation by utilizing displacement constraints, achieving high precision and artifact-free results. The smoothness of the calibrated image is significantly better than that of traditional methods. Secondly, the imaging self-calibration scheme provided by this invention eliminates the need to purchase film or water tanks, fully utilizing existing radiotherapy equipment. Furthermore, the process can be automatically executed by the treatment bed under software control, requiring no consumables and being fully automated. This greatly reduces equipment calibration costs and time, and is applicable to all two-dimensional radiotherapy detectors based on integral modes (EPID, matrix ionization chambers, scintillators, etc.), demonstrating strong equipment versatility and universality. Attached Figure Description

[0011] Figure 1 This is a schematic flowchart of the imaging self-calibration method based on multi-scale displacement constraints provided by the present invention.

[0012] Figure 2 This is a schematic diagram of the image acquisition process based on multi-scale displacement constraints provided by the present invention.

[0013] Figure 3 This is a schematic diagram of fixed displacement sampling provided by the present invention.

[0014] Figure 4 This is a schematic diagram of multi-scale displacement sampling provided by the present invention.

[0015] Figure 5 This is a schematic diagram of the multi-scale global matrix reconstruction process provided by the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application.

[0017] In the description of this invention, the terms "center," "longitudinal," "lateral," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the scope of protection of this invention.

[0018] Example 1 This invention provides a high-precision self-calibration method that can be fully automated by acquiring images using only multi-scale constrained motion, without requiring external reference sources such as film or water tanks, or an absolutely flat accelerator source. By introducing multi-scale displacement to construct a fully connected constraint network and using the global sparse matrix least squares method, a high-precision beam distribution and detector gain matrix are calculated, thereby achieving the purpose of flat-field correction for electronic imaging devices.

[0019] Taking coprime displacement as an example, refer to Figure 1 The paper presents a flowchart of an imaging self-calibration method based on multi-scale coprime displacement constraints, which includes the following steps: Step S01: Acquire a reference image and simultaneously acquire images based on at least two coprime displacements to obtain a beam image sequence containing the reference image and displacement image sequences.

[0020] In this embodiment, the first step is to use a detector to acquire multi-scale displacement-constrained beam images of the beam emitted by the accelerator. In a preferred embodiment, the detector is fixed on a high-precision treatment bed, such as... Figure 2 As shown, the beam source is stationary. After activation, it emits a beam with the direction opposite to the front of the detector. The beam is received and detected by the detector. With the center of the detector as the origin, the plane where the detector is located is the XY plane coordinate system.

[0021] When image acquisition begins, the detector is positioned at the center of the beam by controlling the movement of the treatment bed, at which point a reference image is acquired. (i.e., displacement) Then, by controlling the treatment bed to move along the X and Y axes with at least two sets of coprime sampled displacements based on the beam center, multiple sets of corresponding displacement image sets are obtained. For example, in one example, the displacement is... =2 (in all embodiments, the unit of displacement is the number of pixels). The first set of displacement images I1{} were acquired in the positive and negative directions of the X / Y axes, and the displacement was taken as... =3 The second set of displacement images I2{} were acquired in the positive and negative directions of the X / Y axes respectively. In another example, the displacement is... =2. The first set of displacement images I1{} was acquired in the positive X / Y direction, and the displacement was used as... =5 Simultaneously acquires the first set of displacement images I2{} in the negative X / Y directions. This embodiment of the invention does not impose specific limitations on this.

[0022] In a preferred embodiment, the high-precision treatment bed built into the medical linear accelerator can be used to carry the radiotherapy imaging equipment for image acquisition, and the present invention does not specifically limit this.

[0023] In a preferred embodiment, in order to achieve sampling under different displacements, in addition to moving the detector, the beam field can also be moved by adjusting the accelerator collimator or the multi-leaf grating. The present invention does not specifically limit this.

[0024] In a preferred embodiment, the sampling displacement is required to be coprime between displacements in a sampling direction. For example, and They are numerically coprime (their greatest common divisor is 1). This is also the basis for constructing fully connected constrained networks.

[0025] By introducing two or more sets of coprime displacements in the sampling step to achieve multi-scale coprime displacement constraints, a fully connected graph between pixels can be constructed at the algebraic level, avoiding the "jagged" artifact problem caused by pixel region decoupling. This is also the core of artifact elimination in the embodiments of this invention. Figure 3 As shown, if only the existing single fixed step size (such as displacement) is used... The subsequent equations can only be established and The connection between them. This will cause the entire pixel space to split into two unrelated subnets (even-numbered grids and odd-numbered grids). Due to the lack of coupling constraints between the subnets, the calculated beam intensity Systematic biases can easily occur between subnets, manifesting as high-frequency "comb-like" or "sawtooth" artifacts. To avoid this problem, embodiments of the present invention employ a constraint coupling mechanism based on a multi-scale displacement constraint strategy, such as... Figure 4 As shown, the displacement is respectively 2 and If image acquisition is performed using a network with pixel = 3, then all pixels can be connected via a cross-connection network. For example, pixel 0 → pixel 3 can be connected via a cross-connection network. The pixels are connected, while pixel 3 and pixel 1 are constrained by other paths. It can be seen that the coprime displacements fulfill Bézout's theorem, that is, for any two pixels... and There exists a path through a finite number of times and By combining the connected paths, the solution values ​​of all pixels are tightly "locked" in a unified solution space. Any local deviation will be pulled back by the global grid, thus completely eliminating jagged artifacts in principle. A fully connected graph is formed on the entire pixel plane and used for subsequent multi-scale global matrix construction.

[0026] In a preferred embodiment, in addition to regular coprime displacements, completely random dithering displacements can also be used for image sampling. Since the purpose of this embodiment is to construct a fully connected matrix, and completely random displacements often involve fractional-level sub-pixel displacements, or each displacement does not exhibit a greatest common divisor pattern, completely random dithering displacements will also completely break any fixed spatial intervals, allowing the equations generated by each overlap to intricately interweave and link pixel nodes at different positions across the entire field. With a sufficient number of equations, a mathematically natural connected network covering the entire image can be formed, thereby eliminating artifacts.

[0027] Step S2: Construct a logarithmic domain difference physical model, and use the logarithmic domain difference model to transform the acquired beam image sequence to obtain the logarithmic distribution of beam intensity difference corresponding to pixel displacement.

[0028] To improve the efficiency and accuracy of two-dimensional radiotherapy imaging by utilizing sparse reconstruction in the number domain, we first construct... The logarithmic domain difference model is used to eliminate unknown gain terms, and a sparse matrix of the beam signal is generated based on the difference model and then reconstructed.

[0029] Specifically, due to radiotherapy imaging equipment Gray values ​​collected at the location From the intensity of the incident beam and detector gain The factors are jointly determined, therefore its physical model is a multiplicative noise model, which can be expressed as: (1) To transform the nonlinear multiplication problem into a linear addition problem, taking the natural logarithm of both sides of equation (1) yields: (2) When the detector is displaced Time (i.e., the movement of the detector relative to the beam) For the same location on the detector The gain of the physical pixel at that location It remains unchanged, but the observed beam position changes. The observation equation at this point is expressed as: (3) The differential operation is used to subtract the physical pixels at corresponding positions in the displacement image from those in the reference image, thus eliminating unknown gain terms. get: (4) Let the logarithmic value of the true beam intensity to be solved be Let the logarithmic difference of the detected beam intensity be . Then, the above equation (4) is transformed into the following linear constraint: (5) The physical meaning of formula (5) is: even if the location is unknown. and The absolute beam intensity at a given location can be determined, but by comparing the two measurements, the relative beam intensity difference between the two locations before and after the displacement can be accurately obtained.

[0030] Where k represents the displacement sequence, when there is only one sampling displacement. It can be directly abbreviated as .

[0031] After obtaining the logarithmic domain difference model, it can be used to transform the acquired beam image, linearly mapping the image pixel grayscale values ​​to obtain the result related to the displacement. The corresponding logarithmic distribution of beam intensity difference.

[0032] In a preferred embodiment, the acquired beam image can be preprocessed before conversion to remove obvious noise from the acquired image.

[0033] Step S3: Construct a fully connected global matrix based on multi-scale displacement constraints and perform reconstruction and restoration.

[0034] In this embodiment of the invention, in order to construct a multi-scale global matrix, instead of processing each beam image independently, each effective overlapping pixel pair in all M beam images is transformed into a linear equation, which is then stacked together into a huge sparse matrix equation system. .

[0035] Where M is the total number of beam images actually acquired, determined by the preset number of displacement image groups N. For example, when M=4N+1, corresponding to the image acquisition movement method in the aforementioned embodiment, 4 represents the four directions (positive and negative) on the X / Y axis, and 1 represents the reference image. "Effective overlapping pixel pairs" refer to the field distribution that, although the imaging position changes after the imaging device undergoes physical displacement, it can still capture a portion that overlaps with the reference image. Effective overlapping pixel pairs are pixel pairs between the displacement image and the reference image that are within the beam range and at the same detector position. In the subsequent processing of this embodiment, only these effective overlapping pixel pairs are used to establish an accurate system of equations. .

[0036] Because the two-dimensional global matrix of the entire graph is directly constructed. This can lead to an exponential explosion in computational load (e.g.) The image of pixels contains four million parameters to be solved. This embodiment of the invention employs a dimensionality reduction strategy for step-by-step computation, such as... Figure 5 As shown, the two-dimensional problem is decomposed into multiple one-dimensional solutions in two one-dimensional dimensions, and the computational complexity is reduced by several orders of magnitude by passing through the benchmark anchor point while ensuring global consistency, thus achieving efficient computation.

[0037] Taking "Y-axis baseline + X-axis expansion" as an example, specifically, the construction of a multi-scale global matrix based on dimensionality reduction includes the following steps: Step S301: Determine the center column as the vertical reference on the Y-axis, construct a one-dimensional linear equation system for the center column and solve it to obtain the true relative beam intensity value corresponding to each pixel in the center column.

[0038] The equation system is constructed as follows: Extract the reference image and all displacement images that have moved along the Y-axis. Obtain the logarithmic pixel grayscale values ​​of the central column (Y-axis) corresponding to the geometric center position. Construct only a one-dimensional linear equation system for this central column. The constructed one-dimensional linear equation system is essentially a standard large-scale overdetermined sparse linear equation system, with the following algebraic form: .

[0039] For example, the above steps are described in detail below with reference to specific embodiments. For ease of description, it is assumed that the center column of the detector is extremely simplified to only 5 pixels, with pixel indices [0,1,2,3,4,]. The unknown and to be determined are the logarithmic values ​​of the actual beam intensity at these 5 positions, denoted as the column vector. .

[0040] Assume a reference image was acquired during the sampling step. (i.e., displacement) ), and displacement in the Y-axis direction and displacement Two sets of coprime displacement images were collected.

[0041] (a) Regarding displacement Image conversion: In step S01, the detector acquired a physical displacement of 2 pixels. After eliminating the detector's own gain term, the logarithmic difference of the gray levels at the corresponding pixel positions (denoted as the known observation constant) was calculated. This precisely corresponds to the logarithmic difference in beam intensity between beams spaced 2 pixels apart: For physical pixels

[0042] For physical pixels

[0043] For physical pixels

[0044] (b) Targeting Image conversion: Similarly, the constraint with a span of 3 is generated for a physical displacement of 3 pixels as follows: For physical pixels

[0045] For physical pixels

[0046] (c) Introduce a first reference anchor point constraint as an absolute reference: To eliminate the rank deficiency of the equation system (i.e., the problem that relative values ​​cannot determine the absolute reference), this embodiment of the invention forcibly sets the relative beam intensity of the center pixel (corresponding to index 2) to 1, that is, the logarithmic value is 0, expressed as: The corresponding first reference anchor point constraint is expressed as: .

[0047] Combining all the above constraints, we can transform the system into matrix form. It is expressed as follows: (6) As mentioned earlier, in practical applications, such as using a detector with a width of 2000 pixels for image acquisition and conversion, the coefficient matrix... It will be a highly sparse matrix, with each row containing only two non-zero elements (i.e., two non-zero elements respectively). and Furthermore, the total number of equations (total number of rows, i.e., the sum of all valid overlapping pixel pairs) is much larger than the number of unknowns. The number of columns (total number of columns) constitutes an "overdetermined" system of equations with extremely strong noise resistance and robustness. Coprime step sizes are interwoven in the matrix. In this process, the solution space is completely locked mathematically, and finally, the unique beam intensity distribution column vector y can be obtained by using sparse matrix solving algorithms such as LSQR.

[0048] In a preferred embodiment, the constraint of the first reference anchor point here can also be: If any one of them is 0, it is sufficient to eliminate the rank deficiency of the equation system and reconstruct the beam intensity distribution column vector of the displacement.

[0049] Solving this system of one-dimensional linear equations (6), the central column obtained is denoted as... The central column here. Instead of the pixel grayscale values ​​of the original image, it is a calculated one-dimensional array (vector) that represents the relative intensity distribution curve of the beam on the center line of the Y-axis, providing absolute reference anchor point values ​​for each row of data expanded to both sides of the X-axis in the next step.

[0050] Therefore, in this process, the embodiments of the present invention introduce a first reference anchor point: setting the center point of the center column. (i.e., relative strength is 1), eliminate the rank deficiency of the system of equations, and ensure the uniqueness of the solution to the system of equations.

[0051] In a preferred embodiment, the conjugate gradient method or algebraic reconstruction technique can also be used to solve the one-dimensional linear equation system.

[0052] Step S302: Using each point on the center column as the second reference anchor point and the solved relative beam intensity value of each point on the center column as the value of the second reference anchor point, construct a one-dimensional linear equation system corresponding to each row in the X-axis direction until all rows are traversed; solve the one-dimensional linear equation system corresponding to each row to obtain the true relative beam intensity distribution corresponding to the image pixel position.

[0053] In this embodiment of the invention, the center column obtained in the previous step... Using the second reference anchor point, extend and calculate each row. Specifically, for the first row of the image... rows (total) 1 pixel, unknown number arrive Extract all displacement images that have undergone displacement in the X-axis direction and the reference image at the 1st... For the data in the row, use the same general equation-setting method as in the previous step to construct an equation of the form... A system of one-dimensional equations.

[0054] Because the center point of the center column was set in the previous step when calculating the Y-axis. (Relative value 1). However, if no constraints are applied when calculating each row, the calculated brightness reference for each row will be different. Therefore, in solving the first row, the embodiment of the present invention... When dealing with the overdetermined system of equations, an absolute multi-scale global matrix equation based on the logarithmic distribution of beam intensity difference for effectively overlapping pixel pairs is added: This is enforced by setting the j-th row and the center column... The value of the intersecting pixel (i.e., the center point of the j-th row) is equal to the logarithmic value of the relative beam intensity of that point in the corresponding center column, calculated in the previous step. (7) By iterating through the above forced assignment process line by line ( Until the last row), since the equation for each row independently solves for the true relative beam intensity value of a row of pixels, when the loop ends, the true relative beam intensity value of every pixel in the entire two-dimensional image has been solved, resulting in the true relative beam intensity distribution corresponding to the two-dimensional image. This is equivalent to obtaining the center point of each row by solving the Y-axis center column as described above, thereby precisely constraining each row.

[0055] In a preferred embodiment, for the sake of clarity, it is still assumed that the detector's first... The row is extremely simplified to 5 pixels, with the pixel index as follows: At this time, the index of the corresponding center pixel is . The unknown and undetermined true beam intensity logarithmic value is denoted as a vector. .

[0056] Assuming that a displacement image has been acquired along the X-axis in step S01, and the displacements are respectively... and Construct a system of linear equations using the same method as in the previous step: (a) Targeting : For physical pixels :

[0057] For physical pixels :

[0058] For physical pixels :

[0059] (b) Regarding : For physical pixels :

[0060] For physical pixels :

[0061] (c) Introduce a second reference anchor point as a constraint: In solving the... When dealing with the overdetermined system of equations for a row, add a constraint equation for a second reference anchor point: that is, set the center pixel point (index 1) where the j-th row intersects the center column. The value of ) is equal to the value of the central column obtained in the previous step. The logarithmic value of the relative intensity of the beam at each point, i.e.:

[0062] In summary, the first Transform the system of equations into matrix form It can be expressed as the following formula: (8) As can be seen, the last row of the constant term on the right side of matrix equation (8) is precisely the benchmark value of the central column calculated and passed from the previous step. In this way, the computational complexity of the entire multi-scale global matrix solution is reduced from... Down to This approach avoids massive two-dimensional matrix operations while ensuring that the reconstructed two-dimensional surface is geometrically continuous and consistent. Assume that M samples are obtained. For a detector image with 4,000,000 pixels, the intensity of 4,000,000 unknown pixels needs to be determined. Using a direct two-dimensional global solution without dimensionality reduction will result in an unknown vector of length 4 million and a matrix... The scale is as high as tens of millions of lines 4 million columns. The computational complexity of the least squares method for solving a matrix of this size is approximately... This method requires enormous memory resources and can take days and nights, causing computers or servers to crash due to OutOfMemory (OOM). However, according to the improved split-axis calculation scheme in this invention embodiment, in the first step of the Y-axis calculation, only 2000 unknowns are solved, resulting in a small matrix of 2000 columns, and the calculation is completed instantly. In the second step of the X-axis calculation, each row is solved independently, i.e., 2000 unknowns are solved. Through a simple "for j in range(2000)" loop, 2000 such small one-dimensional matrices are solved sequentially, with the complexity of each one-dimensional solution being only [missing information]. Therefore, the overall computational load increases from non-computable levels. Rapidly descending This allows for the accurate reconstruction of the entire 4-megapixel field within tens of seconds on a regular industrial control computer.

[0063] It should be noted that the above embodiment is implemented by first finding the center column along the Y-axis and then expanding along the X-axis to find each row. In another preferred embodiment, it can also be implemented by first finding the center row along the X-axis and then expanding along the Y-axis to find each column, as long as the two-dimensional problem can be decomposed into multiple one-dimensional solutions in two one-dimensional dimensions.

[0064] Step S04: Obtain the gain matrix using the real relative beam intensity distribution and reference image, and then correct each subsequent acquired beam image to be corrected to obtain the calibrated image.

[0065] The aforementioned full-field reconstruction steps are used to obtain the true relative beam intensity distribution corresponding to the image pixel positions. Then, substitute it into the reference image. The gain matrix of the detector can be obtained by inverse solving. : (9) Then, for any randomly acquired raw beam image to be corrected after subtracting the background... Using the gain matrix The flat-field correction can be completed by performing the following calculation to obtain the calibrated image. : (10).

[0066] This embodiment constructs a fully connected graph between pixels through displacement constraints, avoiding the "jagged" artifact problem caused by pixel region decoupling. At the same time, it uses a split-axis dimension reduction and anchoring strategy to creatively decompose the two-dimensional large-scale reconstruction into two one-dimensional processes with one dimension as the baseline and the other dimension as the expansion. By passing the anchor point, the computational complexity is reduced by several orders of magnitude while ensuring global consistency, thus achieving efficient computation.

[0067] Example 2 This embodiment is an imaging self-calibration system based on multi-scale displacement constraints. The system is used to implement any of the aforementioned imaging self-calibration methods based on multi-scale displacement constraints. The system includes the following modules: The data acquisition module is used to acquire a reference image and perform image acquisition based on at least two sampling displacements to obtain a beam image sequence containing the reference image and the displacement image; The model building module is used to build a logarithmic domain difference model and use the logarithmic domain difference model to transform the beam image to obtain the logarithmic distribution of the beam intensity difference corresponding to the pixel displacement. The split-axis dimensionality reduction and reconstruction module is used to construct a dimensionality-reduced multi-scale global matrix for effectively overlapping pixels based on the logarithmic distribution of beam intensity difference, and to reconstruct and restore the dimensionality-reduced multi-scale global matrix to obtain the true beam intensity distribution corresponding to the image pixel position. The self-calibration module obtains the gain matrix using the real relative beam intensity distribution and reference image, and then calibrates the beam image to be calibrated to obtain the calibration image.

[0068] Example 3 Another embodiment of the present invention provides a corresponding electronic device. The electronic device can be manifested as a general-purpose computing device, such as a server device. The components of the electronic device may include, but are not limited to: at least one processor, at least one memory, and a bus connecting different system components (including the memory and the processor). The bus includes a data bus, an address bus, and a control bus.

[0069] The memory may include volatile memory, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).

[0070] The memory may also include program tools having at least one program module, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0071] The processor executes various functional applications and data processing, such as the methods provided in any of the above embodiments, by running computer programs stored in memory.

[0072] Electronic devices can also communicate with one or more external devices (such as a keyboard). This communication can be achieved through input / output (I / O) interfaces. The electronic devices corresponding to the model can also communicate with one or more networks via network adapters. As shown in the figure, the network adapter communicates with other modules of the electronic devices generated from the model via a bus. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic devices generated from the model, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0073] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0074] Example 4 This embodiment provides a non-transitory computer-readable storage medium storing computer instructions thereon, the computer instructions being used to cause a computer to implement the method provided in any of the above embodiments.

[0075] The present invention has been described in detail above with reference to the accompanying drawings. However, it should be noted that the examples described above are merely preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of the claims of the present invention.

[0076] All features disclosed in this specification, or steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps. This invention is not limited to the specific embodiments described above. This invention extends to any new feature or any new combination disclosed in this specification, as well as any new step or any new combination of any disclosed method or process.

Claims

1. An imaging self-calibration method based on multi-scale displacement constraints, characterized in that, Includes the following steps: Acquire a reference image and perform image acquisition based on the sampling displacement of at least two constraints to obtain a beam image sequence containing the reference image and the displacement image; A logarithmic difference model is constructed, and the beam image sequence is transformed using the logarithmic difference model to obtain the logarithmic distribution of the beam intensity difference corresponding to the displacement. Based on the logarithmic distribution of beam intensity difference and effective overlapping pixel pairs, a dimension-reduced multi-scale global matrix is ​​constructed, and the dimension-reduced multi-scale global matrix is ​​reconstructed and restored to obtain the true relative beam intensity distribution. The gain matrix is ​​obtained by using the real relative beam intensity distribution and the reference image, and the beam image to be calibrated is then calibrated to obtain the calibration image.

2. The imaging self-calibration method based on multi-scale displacement constraints as described in claim 1, characterized in that, The constrained sampling displacement refers to either the sampling displacements in a sampling direction being coprime, or the sampling displacement being a completely random jittery displacement.

3. The imaging self-calibration method based on multi-scale displacement constraints as described in claim 2, characterized in that, The logarithmic domain difference model converts the relative beam intensity difference between two location points where displacement occurs into the difference of beam intensity detection values.

4. The imaging self-calibration method based on multi-scale displacement constraints as described in claim 1, characterized in that, The effective overlapping pixel pair is a pair of pixels between the displacement image and the reference image that are within the beam range and at the same detector position.

5. The imaging self-calibration method based on multi-scale displacement constraints as described in claim 4, characterized in that, A dimension-reduced multi-scale global matrix is ​​constructed based on the logarithmic distribution of beam intensity difference and effective overlapping pixel pairs. include, Based on the first reference anchor point constraint and the logarithmic distribution of the beam intensity difference in the first dimension, the relative beam intensity distribution value corresponding to the center vector of the first dimension is solved; Using the relative intensity distribution of the beam as the second reference anchor point constraint condition of the second dimension, all vectors of the second dimension are solved to obtain a dimension-reduced multi-scale global matrix, which is associated with effective overlapping pixel pairs.

6. The imaging self-calibration method based on multi-scale displacement constraints as described in claim 5, characterized in that, The first dimension is the X-axis and the second dimension is the Y-axis; or the first dimension is the Y-axis and the second dimension is the X-axis.

7. An imaging self-calibration method based on multi-scale displacement constraints as described in claim 5 or 6, characterized in that, The vector is a row or column corresponding to the first or second dimension.

8. The imaging self-calibration method based on multi-scale displacement constraints as described in claim 5, characterized in that, The first reference anchor point constraint condition refers to setting the logarithmic value of any point in the first dimension center vector to a fixed value.

9. The imaging self-calibration method based on multi-scale displacement constraints as described in claim 8, characterized in that, The fixed value is 0.

10. An imaging self-calibration method based on multi-scale displacement constraints as described in claim 6 or 7, characterized in that, Using the relative intensity distribution of the beam as the second reference anchor point constraint in the second dimension, solve for all vectors in the second dimension, including: Determine the intersection point of each vector in the second dimension with the central vector in the first dimension; The value of the intersection point is set to the logarithm of the relative intensity of the beam at the corresponding position of the intersection point in the first dimension center vector, thereby associating and constraining each vector of the second dimension with the relative intensity distribution value of the beam. Solve for each vector in the second dimension to obtain the true relative beam intensity distribution corresponding to the pixel position.

11. An imaging self-calibration system based on multi-scale displacement constraints, characterized in that, The system is used to implement the imaging self-calibration method based on multi-scale displacement constraints as described in any one of claims 1-10, and includes the following modules: The data acquisition module is used to acquire a reference image and perform image acquisition based on the sampling displacement of at least two constraints to obtain a beam image sequence containing the reference image and the displacement image; The model building module is used to build a logarithmic domain difference model and use the logarithmic domain difference model to transform the beam image to obtain the logarithmic distribution of the beam intensity difference corresponding to the pixel displacement. The split-axis dimensionality reduction and reconstruction module is used to construct a dimensionality-reduced multi-scale global matrix based on the logarithmic distribution of beam intensity difference and effective overlapping pixel pairs, and to reconstruct and restore the dimensionality-reduced multi-scale global matrix to obtain the true relative beam intensity distribution. The self-calibration module obtains the gain matrix using the real relative beam intensity distribution and reference image, and then calibrates the beam image to be calibrated to obtain the calibration image.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the imaging self-calibration method based on multi-scale displacement constraints as described in any one of claims 1 to 10.

13. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to implement the imaging self-calibration method based on multi-scale displacement constraints as described in any one of claims 1-10.

Images