Laser interference coding image restoration optimization method

Abstract

The invention provides a laser interference coding image restoration optimization method, and belongs to the technical field of photoelectric imaging in image processing. According to the method, comprehensive optimization of image quality is realized through directional search and removal of light spot noise, restoration of coding information and restoration of background information.

Description

Technical Field

[0001] This invention belongs to the field of optoelectronic imaging technology in image processing, and particularly relates to a laser interference coded image restoration optimization method. Background Technology

[0002] Since its inception, laser technology has been widely used in military, industrial, and medical fields due to its excellent directionality, coherence, and high energy density. However, the concentrated energy of high-power lasers poses a serious threat to photoelectric imaging detectors. Especially with the continuous development of laser damage devices, how to protect photoelectric imaging systems from laser damage has become an urgent problem to be solved.

[0003] Currently, common laser protection methods mainly include: optical limiters based on nonlinear optical materials, optical limiters based on coatings and phase change materials, and optical system optimization techniques based on wavefront modulation. Among these, methods based on nonlinear optical materials utilize intensity-dependent absorption characteristics to absorb or scatter excess laser energy during high-power laser irradiation, thus reducing damage to the detector. This method is simple to implement but has poor adaptability to broadband light sources. Optical limiters based on coatings and phase change materials utilize the phase change properties or self-healing mechanisms of materials to dynamically adjust optical properties during strong laser irradiation, blocking or reflecting laser energy to protect the detector. However, these methods have low tolerance to continuous laser interference and poor reusability. Optical system optimization techniques based on wavefront modulation, by altering the laser wavefront morphology, disperse laser energy over a wider range, thereby reducing local energy density. Simultaneously, combined with backend digital signal processing, background image information is restored. This technology not only improves the system's damage resistance but also ensures normal imaging quality, gradually becoming an important development direction in the field of laser protection. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a restoration and optimization scheme for laser-interference coded images. By directional searching and removal of spot noise, repair of coded information, and restoration of background information, comprehensive optimization of image quality is achieved.

[0005] The first aspect of this invention proposes a laser interference coded image restoration optimization method, the method comprising:

[0006] Step S1: Locate the spot noise region in the laser interference coded image using a search method based on light intensity distribution characteristics, and gradually determine the location and range of the spot region by analyzing the local brightness changes in the image;

[0007] Step S2: Remove the located light spot by intensity information matching. Compare the pixel intensity of the light spot area with the surrounding area, remove the light spot according to the comparison result, and retain the background detail information.

[0008] Step S3: Repair and reconstruct the spot area by gradually filling it in. Based on the local texture and structural information of the image, the spot area is gradually restored to the background image.

[0009] Step S4: Based on the PSF model of the vortex optical system, Wiener filtering technology is used to deconvolve the repaired image to restore background details and encoded information.

[0010] In step S1, the laser interference coded image is initially scanned to identify potential high-brightness areas. By combining local contrast and brightness gradient information, the potential high-brightness areas are further verified to determine the boundary and range of the light spot.

[0011] Step S1 specifically includes:

[0012] Step S1(a): Obtain the laser interference coded image and the spot feature template image, define the upper and lower bounds of the search area so that the spot positioning range covers the effective area of ​​the image; set parameters, including population size, maximum number of iterations, and initial value of adaptive probability;

[0013] Step S1(b): Randomly generate a set of candidate solutions, each candidate solution representing a possible spot area location, with a rectangle representing the possible spot area location, each candidate solution including the upper left corner coordinates and the search area size;

[0014] Step S1(c): For the search region of each candidate solution, the similarity between the spot feature template image and the region is calculated using the normalized product similarity matching mechanism, and the region with the highest similarity is taken as the current best solution;

[0015] Step S1(d): In the initial stage, a global search is performed with a first probability to obtain possible spot locations; the spot is located in the entire image by adjusting the position and size of the candidate solutions; after the global search locks a certain region, the local search probability is gradually increased to a second probability to further optimize the spot position;

[0016] Step S1(e): Based on the search results of each iteration, update the coordinates and size of the light spot position; if the similarity of the new solution is better than the current best solution, then replace it with the new solution as the current best solution;

[0017] Step S1(d): Check if the maximum number of iterations has been reached, or if the similarity change of the solution is less than the set threshold; if the termination condition is met, output the determined spot area position.

[0018] Step S2 specifically includes:

[0019] Step S2(a): Based on the spot localization result, mark the detected spot area in the form of a mask, compare the pixel value of the spot area with the intensity information of the surrounding background, and determine the boundary of the spot;

[0020] Step S2(b): Perform statistical analysis on the pixel values ​​within the marked area, including calculating the pixel intensity distribution of the spot area, comparing the pixel intensity of the spot area with the average value of the surrounding background pixels, and determining the threshold range of the spot area based on the comparison results;

[0021] Step S2(c): Based on the pixel distribution characteristics of the spot area, set an intensity threshold T to distinguish spot pixels from background pixels; if the spot pixel value is higher than the threshold T, the pixel is considered to belong to the spot; if the spot pixel value is not higher than the threshold T, the pixel is retained as background information.

[0022] Step S2(d): Set the pixel values ​​in the spot area that exceed the intensity threshold T to zero to remove strong interference signals from the spot, and keep the original intensity values ​​of background pixels outside the marked area to ensure the integrity of the background information;

[0023] Step S2(e): Smooth the boundary between the spot area and the background information to avoid discontinuities or abruptness in the boundary caused by directly assigning zeros. Use Gaussian blur or interpolation methods to smooth the boundary transition area.

[0024] In step S3, based on the pixel information of the edge of the spot area, the neighboring area with the highest similarity is selected for matching, and the matched image blocks are copied and filled into the spot area in turn, thereby gradually completing the repair.

[0025] Step S3 specifically includes:

[0026] Step S3(a): Calculate the priority; mark the area to be repaired Ω, and use the Criminisi algorithm to select an N*N pixel block Ψ centered at point p on the edge δΩ of the area to be repaired. p Pixel block Ψ p The formula for calculating the priority P(p) is:

[0027] P(p) = C(p)·D(p)

[0028] Where C(p) is the confidence term and D(p) is the data term, defined as follows:

[0029]

[0030] Among them, |Ψ p | is a pixel block Ψ p The total number of pixels; ∝ is the normalization factor;

[0031] Step S3(b): Find the best matching block and fill it; using the block to be repaired with the highest priority as the target block, use a global search method with the SSD matching criterion to find the most similar matching block Ψ to the target block. q The patch block is then filled; the SSD matching criterion calculation formula is:

[0032]

[0033] Where I1(i) and I2(i) represent the gray values ​​of the original image and the matching image at the i-th pixel position, respectively, and N is the total number of pixels;

[0034] Step S3(c): Update the confidence score; after filling, the region is included in the intact region category for the next round of calculation. The confidence score of the filled region is expressed as:

[0035]

[0036] Where C(p) is the confidence level of the filled region, and C(q) is the confidence level of the matching pixel block region;

[0037] Step S3(d): Repeat steps S3(a)-S3(c) until the repair is complete.

[0038] Step S4 specifically includes:

[0039] Step S4(a): Construct the imaging model. The observed image g(x,y) is the result of convolving the original image f(x,y) with the system PSFh(x,y) and adding noise n(x,y).

[0040]

[0041] Step S4(b): Convert the image to the frequency domain by using Fast Fourier Transform (FFT) to transform the spatial domain image expression into a frequency domain expression.

[0042] G(u,v)=F(u,v)H(u,v)+N(u,v)

[0043] The frequency domain expression of the original image is obtained as follows:

[0044] F(u,v)=G(u,v)W(u,v)

[0045] Step S4(c): Construct the Wiener filter and define its expression:

[0046]

[0047] Step S4(d): Frequency domain image restoration, applying the Wiener filter to the frequency domain representation of the observed image:

[0048] F(u,v)=G(u,v)W(u,v)

[0049] F(u,v) is the frequency domain representation of the restored original image; the inverse fast Fourier transform (IFFT) is used to convert the restored image back from the frequency domain to the spatial domain:

[0050] f(x,y)=IFFT(F(u,v))

[0051] Step S4(e): Construct boundary condition processing to process the cutoff frequency of the optical system; when H(u,v)=0, the Wiener filter is invalid and the signal at the corresponding frequency cannot be recovered. The blank area in the frequency domain is processed by interpolation or other frequency filling methods to reduce image degradation caused by missing information.

[0052] A second aspect of the present invention provides a laser interference coded image restoration optimization system, the system comprising a processing unit configured to perform:

[0053] A search method based on light intensity distribution characteristics is used to locate the spot noise region in the laser interference coded image. The location and range of the spot region are gradually determined by analyzing the local brightness changes in the image.

[0054] The method of intensity information matching is used to remove the located light spots. The pixel intensity of the light spot area is compared with the surrounding area, and the light spot is removed according to the comparison result, while the background detail information is preserved.

[0055] The spot area is repaired and reconstructed by gradually filling in the image. Based on the local texture and structural information of the image, the spot area is gradually restored to the background image.

[0056] Based on the PSF model of the vortex optical system, Wiener filtering technology is used to perform deconvolution processing on the repaired image, thereby restoring background details and encoded information.

[0057] A third aspect of this invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements a laser interference coded image restoration optimization method according to the first aspect of this disclosure.

[0058] A fourth aspect of this invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements a laser interference coded image restoration optimization method according to the first aspect of this disclosure.

[0059] In summary, this invention systematically optimizes the quality of laser-interference images by combining directional search for spot noise, spot region removal and repair, and background information restoration based on the point spread function (PSF). Through this series of processing steps, this invention not only achieves effective removal of laser spot noise but also successfully utilizes Wiener filtering technology to achieve high-quality image restoration, providing an efficient solution for laser spot repair and analysis in vortex optics systems. Attached Figure Description

[0060] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram of the laser interference coded image restoration optimization process according to an embodiment of the present invention.

[0062] Figure 2 This is a laser interference encoded image according to an embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram of the light spot area according to an embodiment of the present invention.

[0064] Figure 4 This is an intensity threshold mask image according to an embodiment of the present invention.

[0065] Figure 5 This is a schematic diagram of the spot noise removal result according to an embodiment of the present invention.

[0066] Figure 6 This is a schematic diagram of the spot noise restoration result according to an embodiment of the present invention.

[0067] Figure 7 This is a schematic diagram illustrating the background information restoration result according to an embodiment of the present invention.

[0068] Figure 8 This is a schematic diagram of the Levy flight trajectory according to an embodiment of the present invention.

[0069] Figure 9 This is a schematic diagram of the Transformer network architecture according to an embodiment of the present invention.

[0070] Figure 10 This is a schematic diagram illustrating the effect of light spot area restoration according to an embodiment of the present invention.

[0071] Figure 11This is a schematic diagram illustrating the effect of restoring background information according to an embodiment of the present invention. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] First Embodiment

[0074] This embodiment discloses a laser interference coded image restoration optimization method, including the following steps:

[0075] S1: A search method based on light intensity distribution characteristics is used to locate the spot noise region in the laser interference image. By analyzing the local brightness changes in the image, the location and range of the spot region are gradually determined. Specifically, the image is first preliminarily scanned to identify possible high-brightness areas; then, by combining local contrast and brightness gradient information, suspected areas are further verified, thereby accurately locking the boundary and range of the spot, providing support for subsequent processing.

[0076] S2: An intensity information matching method is used to accurately remove light spots. This method analyzes the pixel intensity of the light spot area and compares it with the surrounding area to effectively remove the light spots while preserving background details.

[0077] S3: A gradual filling method is used to repair the light spot area. First, based on the pixel information of the edge of the light spot area, the most similar neighboring region is selected for matching; then, the matched image blocks are sequentially copied and filled into the light spot area, gradually completing the repair. The entire process is based on the local texture and structural information of the image, ensuring that the light spot area is gradually restored to the background image.

[0078] S4: The PSF model based on the vortex optical system uses Wiener filtering technology to perform deconvolution processing on the repaired image to restore background details and encoded information.

[0079] Step S1 specifically includes the following steps:

[0080] S1(a): Obtain the laser interference image to be processed and the template image (spot feature template). Define the upper and lower bounds of the search area to ensure that the spot localization range covers the effective area of ​​the image. Set the algorithm parameters, including population size, maximum number of iterations, and initial value of adaptive probability.

[0081] S1(b): Randomly generate a set of candidate solutions, each candidate solution representing a possible spot location (represented by a rectangle). Each candidate solution includes the coordinates of the top left corner and the size of the search area.

[0082] S1(c): For each candidate solution's search region, calculate the similarity between the template image and the region using a normalized product similarity matching mechanism. Record the region with the highest similarity as the current best solution.

[0083] S1(d): In the initial stage, the algorithm performs a global search with high probability to explore possible spot locations. By adjusting the position and size of candidate solutions, the spot is located in the entire image. Once the global search locks onto a high-similarity region, the probability of the local search is gradually increased to refine the region and further optimize the spot location.

[0084] S1(e): Update the coordinates and size of the light spot position based on the search results of each iteration. If the similarity of the new solution is better than the current best solution, then replace it with the new best solution.

[0085] S1(d): Check if the maximum number of iterations has been reached, or if the similarity change of the solutions is less than a set threshold. If the termination condition is met, output the final determined spot region location.

[0086] Step S2 specifically includes the following steps:

[0087] S2(a): Based on the spot localization results, the detected spot area is marked in the form of a mask. The pixel values ​​of the spot area are initially compared with the intensity information of the surrounding background to determine the boundary of the spot.

[0088] S2(b): Perform statistical analysis on the pixel values ​​within the marked area, including calculating the pixel intensity distribution (such as mean and standard deviation) of the spot area. Compare the pixel intensity of the spot area with the average value of the surrounding background pixels to determine the threshold range of the spot area.

[0089] S2(c): Based on the pixel distribution characteristics of the spot area, an intensity threshold T is set to distinguish spot pixels from background pixels. If the spot pixel value I(x,y) is higher than the threshold T, the pixel is considered to belong to the spot. If the spot pixel value I(x,y) is lower than or close to the threshold T, the pixel is retained as background information.

[0090] S2(d): Set the pixel values ​​in the spot region that exceed the intensity threshold T to zero to remove strong interference signals from the spot. Maintain the original intensity values ​​for background pixels outside the marked area to ensure the integrity of the background information.

[0091] S2(e): Smooth the boundary between the spot area and the background information to avoid discontinuities or abruptness caused by directly assigning zeros. Use Gaussian blur or interpolation methods to smooth the boundary transition area.

[0092] Step S3 specifically includes the following steps:

[0093] S3(a): Calculate priority. After marking the region to be repaired Ω, the Criminisi algorithm selects an N*N pixel block Ψ centered at point p on the edge δΩ of the region to be repaired. p Pixel block Ψ p The formula for calculating the priority P(p) is:

[0094] P(p) = C(p)·D(p)

[0095] Where C(p) is the confidence term and D(p) is the data term, defined as follows:

[0096]

[0097] Among them |Ψ p | is a pixel block Ψ p The total number of pixels; ∝ is the normalization factor, which is usually taken as 255.

[0098] S3(b): Find the best matching block and fill it. After the priority of the blocks to be repaired is calculated, the block with the highest priority is selected as the best matching block Ψ. p As the target block, a global search method is used to find matching blocks Ψ using the SSD (Sum of Squared Different) matching criterion. p Most similar matching block Ψ q Then, the patch is filled. The SSD matching criterion calculation formula is:

[0099]

[0100] Where: I1(i) and I2(i) represent the gray values ​​of the original image and the matching image at the i-th pixel position, respectively, and N is the total number of pixels.

[0101] S3(c): Update confidence. After the damaged area is filled, it is included in the intact area category for the next round of calculation. The edges of the damaged area will also be redefined. At this time, the confidence of the filled area can be expressed as:

[0102]

[0103] Where C(p) is the confidence level of the filled region and C(q) is the confidence level of the matched pixel block region.

[0104] S3(d): Repeat steps S3(a)-S3(c) until the repair is complete.

[0105] Step S4 specifically includes the following steps:

[0106] S4(a): Construct the imaging model, assuming that the observed image g(x,y) is the result of convolving the original image f(x,y) with the system PSFh(x,y) and adding noise n(x,y):

[0107]

[0108] S4(b): Transforming the image to the frequency domain: Use Fast Fourier Transform (FFT) to convert the spatial domain image representation to the frequency domain representation.

[0109] G(u,v)=F(u,v)H(u,v)+N(u,v)

[0110] Simplifying the formula yields the frequency domain expression of the original image:

[0111] F(u,v)=G(u,v)W(u,v)

[0112] S4(c): Construct the Wiener filter and define its expression:

[0113]

[0114] S4(d): Frequency domain image restoration, applying the Wiener filter to the frequency domain representation of the observed image:

[0115] F(u,v)=G(u,v)W(u,v)

[0116] F(u,v) is the frequency domain representation of the restored original image. The inverse fast Fourier transform (IFFT) is used to transform the restored image back from the frequency domain to the spatial domain:

[0117] f(x,y)=IFFT(F(u,v))

[0118] S4(e): Constructing boundary condition processing to address the optical system's cutoff frequency (MTF zero point) problem: When H(u,v) = 0, the Wiener filter is ineffective, resulting in the inability to recover the signal at the corresponding frequency. In this case, it is necessary to process the blank areas in the frequency domain using interpolation or other frequency filling methods to minimize image degradation caused by missing information.

[0119] Second Embodiment

[0120] This embodiment proposes an optimized method for laser interference-coded image restoration. For example... Figure 1The diagram shows a schematic of an optimization model for laser interference coded image restoration, where the laser interference coded image is an interference image generated by a vortex-coded optical system. The experimental dataset is derived from simulated laser interference coded images.

[0121] In this embodiment, an improved flower pollination algorithm (FPA) is used to search for and locate light spot noise, the Criminisi algorithm is used to repair the light spot area, and a Wiener filter-based background restoration method is combined to optimize the overall image quality.

[0122] The laser interference coded image restoration optimization method includes the following steps:

[0123] S1: Directional search of the spot noise region.

[0124] Laser interference encoded images, such as Figure 2 As shown. Step S1 specifically includes the following steps:

[0125] S1(a): Initialize the population and parameters

[0126] Obtain the laser interference image to be processed and the target image template (spot feature template).

[0127] Set initial values ​​for population size, maximum number of iterations, and adaptive search probability.

[0128] S1(b): Population generation and spot region matching

[0129] The pollen population is randomly initialized, and each pollen individual represents a possible spot location (represented by a rectangular region).

[0130] Normalized product similarity is used as the matching objective function to calculate the similarity between the template image and the candidate region, and the position with the highest similarity is recorded as the current best solution.

[0131] S1(c): Switching between global and local search

[0132] The weights of global search (Levy flight) and local search are adjusted based on adaptive probability to optimize the position of individuals in the population.

[0133] S1(d): Determine the spot area

[0134] Repeat the iteration until the maximum number of iterations is reached or the similarity change is less than a set threshold, then output the final location information of the light spot region, such as... Figure 3 As shown.

[0135] S2: Spot noise removal.

[0136] Step S2 specifically includes the following steps:

[0137] S2(a): Intensity threshold setting and pixel zeroing. Based on the intensity difference between the pixel values in the spot area and the background area, set the intensity threshold T. The intensity threshold mask is as shown in Figure 4 . If the pixel value I(x, y) > T, it is regarded as spot noise and assigned a value of zero. If the pixel value I(x, y) < T, the original pixel value is retained to protect the background information as much as possible. Traverse the spot area pixel by pixel and process the pixel values in the area according to the above rules.

[0138] S2(b): Boundary smoothing. For the possible abrupt boundary between the area with zeroed spots and the surrounding background, use boundary smoothing technology for transitional processing. Use Gaussian blur to process the boundary area or adopt interpolation methods to optimize the boundary transition to ensure that the removal of the spot area is smooth and natural. The result of operation S2 is as shown in Figure 5 .

[0139] S3: Spot noise repair.

[0140] S3(a): Priority calculation. Mark the missing area after spot removal as the area to be repaired (Ω) and calculate the repair priority. On the boundary δΩ of the area to be repaired, select a repair block Ψ centered at point p p . The priority calculation formula is: P(p) = C(p)D(p)

[0141] where C(p): confidence term, measuring the proportion of known pixels in the repair block area. D(p): data term, using edge information to describe the importance of the structure.

[0142] S3(b): Search for the best-matching block. Using the full-image search technology and taking SSD (Sum of Squared Difference) as the matching criterion, find the candidate block Ψ that best matches the target repair block Ψ p . The matching criterion calculation formula: q where I1(i) and I2(i) are the pixel values of the target block and the candidate block.

[0143] S3(c): Filling and updating the confidence. Use the pixel values of the best-matching block to fill the current repair block to complete partial repair of the spot missing area. Update the confidence of the repaired area and re-divide the boundary δΩ of the area to be repaired.

[0144] Figure 6 S3(d): Repeat the repair until completion. Repeat the priority calculation, matching block search, and filling steps to repair the spot missing area block by block until all areas to be repaired are filled. The result of operation S3 is as shown in Figure 6 .

[0145] S4: Restoration of background information.

[0146] Background information restoration results, such as Figure 7 As shown, step S4 specifically includes the following steps:

[0147] S4(a): Use Fast Fourier Transform (FFT) to convert the image after removing the blemishes from the spatial domain to the frequency domain, and obtain the frequency domain expression of the observed image.

[0148] S4(b): Based on the point spread function (PSF) of the vortex optical system, establish the Wiener filter model: SNR (Signal-to-Noise Ratio) is an indicator used to measure the relationship between signal strength and background noise intensity.

[0149] S4(c): Apply a Wiener filter to the frequency domain image for deconvolution processing to restore the background blur caused by interference. Use an inverse fast Fourier transform (IFFT) to convert the restored image back from the frequency domain to the spatial domain.

[0150] As a preferred embodiment of the present invention, the laser interference coded image restoration optimization method is based on non-neural network image processing technology. The method mainly relies on an improved flower pollination algorithm (FPA) for spot region localization, the Criminisi algorithm for spot region repair, and Wiener filtering technology for background information restoration.

[0151] In this embodiment of the invention, a grayscale image with a resolution of 600×800 was selected to construct a total of 30 experimental data images with laser interference. The experimental data consisted of 20 training images used to optimize the parameters of the FPA and restoration algorithm; and 10 test images used to verify the restoration effect. During the testing phase, the restoration performance of the proposed method was verified by comparing the PSNR and SSIM values ​​of the restored image and the interference-free original image.

[0152] Third Embodiment

[0153] In one embodiment of the present invention, the image data is a visible light image dataset. The image dataset of the present invention originates from images generated by a self-generated vortex-coded optical system, mainly covering scenes such as aircraft.

[0154] In this embodiment, a Transformer neural network based on a self-attention mechanism is used to replace the traditional Criminisi algorithm for repairing the light spot area.

[0155] The laser interference infrared coded image restoration and optimization method of the present invention includes the following steps:

[0156] S1: Directional search of the spot noise region.

[0157] Step S1 specifically includes the following steps:

[0158] S1(a): An infrared image with a resolution of 600×800 is selected as input. The population size of the improved FPA algorithm is set to 40, and the maximum number of iterations is set to 150. The normalized product similarity is used as the objective function to measure the matching degree between the target template and the spot region.

[0159] S1(b): Initialize the population, where each pollen grain represents the top-left corner coordinates of a possible light spot region. In each iteration, the pollen grain position is dynamically updated using a hybrid mechanism of global search (Levy flight) and local search. The Levy flight trajectory is as follows: Figure 8 As shown.

[0160] S1(c): After several iterations, the location of the light spot region is determined, and the coordinates and size of the upper left corner of the light spot region are output.

[0161] S2: Spot noise removal.

[0162] S2(a): Analyze the pixel values ​​within the spot area determined in S1, set an intensity threshold T, and set pixels exceeding the threshold to zero. Use Gaussian blur to smooth the spot boundary to avoid abrupt edges after removing the spot.

[0163] S3: Spot area repair.

[0164] S3(a): Transformer model training. Training and label samples are constructed using a labeled dataset of laser-interferenced infrared images. Sample data: Infrared images with the laser-interferenced regions removed. Label data: Uninterferenced raw infrared images.

[0165] A Transformer neural network is constructed, with a symmetric encoder and decoder. The encoder consists of four convolutional blocks, each composed of two 3×3 convolutional layers and a ReLU activation function, downsampling using 2×2 max-pooling layers. The decoder consists of an upsampling layer (2×2), two 3×3 convolutional layers, and a ReLU activation function, combined with skip connections to pass the high-resolution features from the encoder output to the decoder. The Transformer network architecture is as follows: Figure 9 As shown.

[0166] The mean squared error (MSE) was used as the loss function, and the Adam optimizer was used for training. The training data consisted of 10,000 images, and the number of iterations was 50.

[0167] S3(b): Input the infrared image after removing the blemishes into the trained Transformer model to generate the restored image. Compare the restored image with the interference-free image to ensure the consistency of texture and structure in the restored area. The restoration effect is as follows: Figure 10 As shown.

[0168] S4: Background information restoration.

[0169] Background restoration effect as follows: Figure 11 As shown, step S4 specifically includes the following steps:

[0170] S4(a): Perform Fast Fourier Transform (FFT) on the infrared image after spot restoration to obtain the frequency domain representation.

[0171] S4(b): Constructing a Wiener filter using the point spread function (PSF) of a vortex optics system: Where H(u,v) is the frequency domain representation of the PSF, S n and S f These are noise and signal power, respectively.

[0172] S4(c): Background restoration.

[0173] By applying the Wiener filter to the frequency domain image and using the inverse fast Fourier transform (IFFT) to convert the restored image back to the spatial domain, a clear background image is obtained.

[0174] As a preferred embodiment of the present invention, the method for restoring and optimizing laser-interferenced infrared coded images is based on a combination of deep learning and image processing technologies. The method mainly relies on an improved flower pollination algorithm (FPA) for locating the light spot region, a Transformer neural network for repairing the light spot region, and Wiener filtering technology for restoring background information.

[0175] In this embodiment, 2000 infrared images with a resolution of 600×800 were selected to construct experimental data with laser interference. The experimental data consisted of 1900 training images used to optimize the FPA algorithm and train the Transformer restoration model; and 100 test images used to verify the restoration effect. During the testing phase, the restoration performance of the proposed method was verified by comparing the PSNR and SSIM values ​​of the restored image and the interference-free original image. Experimental results show that the SSIM improved from 0.94 to 0.9945 after restoration using this method, effectively verifying the feasibility and superiority of the method.

[0176] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A laser interference-coded image restoration optimization method, characterized in that, The method includes: Step S1: Locate the spot noise region in the laser interference coded image using a search method based on light intensity distribution characteristics, and gradually determine the location and range of the spot region by analyzing the local brightness changes in the image; Step S2: Remove the located light spot by intensity information matching. Compare the pixel intensity of the light spot area with the surrounding area, remove the light spot according to the comparison result, and retain the background detail information. Step S3: Repair and reconstruct the spot area by gradually filling it in. Based on the local texture and structural information of the image, the spot area is gradually restored to the background image. Step S4: Based on the PSF model of the vortex optical system, Wiener filtering technology is used to deconvolve the repaired image to restore background details and encoded information.

2. The laser interference coded image restoration optimization method according to claim 1, characterized in that, In step S1, the laser interference coded image is initially scanned to identify potential high-brightness areas. By combining local contrast and brightness gradient information, the potential high-brightness areas are further verified to determine the boundary and range of the light spot.

3. The laser interference coded image restoration optimization method according to claim 2, characterized in that, Step S1 specifically includes: Step S1(a): Obtain the laser interference coded image and the spot feature template image, define the upper and lower bounds of the search area so that the spot positioning range covers the effective area of ​​the image; set parameters, including population size, maximum number of iterations, and initial value of adaptive probability; Step S1(b): Randomly generate a set of candidate solutions, each candidate solution representing a possible spot area location, with a rectangle representing the possible spot area location, each candidate solution including the upper left corner coordinates and the search area size; Step S1(c): For the search region of each candidate solution, the similarity between the spot feature template image and the region is calculated using the normalized product similarity matching mechanism, and the region with the highest similarity is taken as the current best solution; Step S1(d): In the initial stage, a global search is performed with a first probability to obtain possible spot locations; the spot is located in the entire image by adjusting the position and size of the candidate solutions; after the global search locks a certain region, the local search probability is gradually increased to a second probability to further optimize the spot position; Step S1(e): Based on the search results of each iteration, update the coordinates and size of the light spot position; if the similarity of the new solution is better than the current best solution, then replace it with the new solution as the current best solution; Step S1(d): Check if the maximum number of iterations has been reached, or if the similarity change of the solution is less than the set threshold; if the termination condition is met, output the determined spot area position.

4. The laser interference coded image restoration optimization method according to claim 3, characterized in that, Step S2 specifically includes: Step S2(a): Based on the spot localization result, mark the detected spot area in the form of a mask, compare the pixel value of the spot area with the intensity information of the surrounding background, and determine the boundary of the spot; Step S2(b): Perform statistical analysis on the pixel values ​​within the marked area, including calculating the pixel intensity distribution of the spot area, comparing the pixel intensity of the spot area with the average value of the surrounding background pixels, and determining the threshold range of the spot area based on the comparison results; Step S2(c): Based on the pixel distribution characteristics of the spot area, set an intensity threshold T to distinguish spot pixels from background pixels; if the spot pixel value is higher than the threshold T, the pixel is considered to belong to the spot; if the spot pixel value is not higher than the threshold T, the pixel is retained as background information. Step S2(d): Set the pixel values ​​in the spot area that exceed the intensity threshold T to zero to remove strong interference signals from the spot, and keep the original intensity values ​​of background pixels outside the marked area to ensure the integrity of the background information; Step S2(e): Smooth the boundary between the spot area and the background information to avoid discontinuities or abruptness in the boundary caused by directly assigning zeros. Use Gaussian blur or interpolation methods to smooth the boundary transition area.

5. The laser interference coded image restoration optimization method according to claim 4, characterized in that, In step S3, based on the pixel information of the edge of the spot area, the neighboring area with the highest similarity is selected for matching, and the matched image blocks are copied and filled into the spot area in turn, thereby gradually completing the repair.

6. The laser interference coded image restoration optimization method according to claim 5, characterized in that, Step S3 specifically includes: Step S3(a): Calculate the priority; mark the area to be repaired Ω, and use the Criminisi algorithm to select an N*N pixel block Ψ centered at point p on the edge δΩ of the area to be repaired. p Pixel block Ψ p The formula for calculating the priority P(p) is: P(p) = C(p)·D(p) Where C(p) is the confidence term and D(p) is the data term, defined as follows: Among them, |Ψ p | is a pixel block Ψ p The total number of pixels; ∝ is the normalization factor; Step S3(b): Find the best matching block and fill it; using the block to be repaired with the highest priority as the target block, use a global search method with the SSD matching criterion to find the most similar matching block Ψ to the target block. q The patch block is then filled; the SSD matching criterion calculation formula is: Where I1(i) and I2(i) represent the gray values ​​of the original image and the matching image at the i-th pixel position, respectively, and N is the total number of pixels; Step S3(c): Update the confidence score; after filling, the region is included in the intact region category for the next round of calculation. The confidence score of the filled region is expressed as: Where C(p) is the confidence level of the filled region, and C(q) is the confidence level of the matching pixel block region; Step S3(d): Repeat steps S3(a)-S3(c) until the repair is complete.

7. The laser interference coded image restoration optimization method according to claim 6, characterized in that, Step S4 specifically includes: Step S4(a): Construct the imaging model. The observed image g(x,y) is the result of convolving the original image f(x,y) with the system PSFh(x,y) and adding noise n(x,y). Step S4(b): Convert the image to the frequency domain by using Fast Fourier Transform (FFT) to transform the spatial domain image expression into a frequency domain expression. G(u,v)=F(u,v)H(u,v)+N(u,v) The frequency domain expression of the original image is obtained as follows: F(u,v)=G(u,v)W(u,v) Step S4(c): Construct the Wiener filter and define its expression: Step S4(d): Frequency domain image restoration, applying the Wiener filter to the frequency domain representation of the observed image: F(u,v)=G(u,v)W(u,v) F(u,v) is the frequency domain representation of the restored original image; the inverse fast Fourier transform (IFFT) is used to convert the restored image back from the frequency domain to the spatial domain: f(x,y)=IFFT(F(u,v)) Step S4(e): Construct boundary condition processing to process the cutoff frequency of the optical system; when H(u,v)=0, the Wiener filter is invalid and the signal at the corresponding frequency cannot be recovered. The blank area in the frequency domain is processed by interpolation or other frequency filling methods to reduce image degradation caused by missing information.

8. A laser interference-encoded image restoration and optimization system, characterized in that, The system includes a processing unit configured to perform: A search method based on light intensity distribution characteristics is used to locate the spot noise region in the laser interference coded image. The location and range of the spot region are gradually determined by analyzing the local brightness changes in the image. The method of intensity information matching is used to remove the located light spots. The pixel intensity of the light spot area is compared with the surrounding area, and the light spot is removed according to the comparison result, while the background detail information is preserved. The spot area is repaired and reconstructed by gradually filling in the image. Based on the local texture and structural information of the image, the spot area is gradually restored to the background image. Based on the PSF model of the vortex optical system, Wiener filtering technology is used to perform deconvolution processing on the repaired image, thereby restoring background details and encoded information.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the laser interference coded image restoration optimization method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the laser interference coded image restoration optimization method according to any one of claims 1-7.

Images