An imaging method

Abstract

This invention belongs to the field of high-resolution image reconstruction technology, specifically relating to an image denoising method and an imaging method. The image denoising method includes: acquiring multiple preprocessed images; dividing the multiple preprocessed images into multiple groups, each group including multiple preprocessed images; for each group of preprocessed images, obtaining a guide map based on all preprocessed images within the group, resulting in multiple guide maps, each corresponding one-to-one with one of the groups of preprocessed images; and using the guide maps corresponding to each group of preprocessed images to perform guided filtering on each preprocessed image within the group. The image denoising method of this application performs guided filtering on the preprocessed images in groups, calculates the guide map for each group of preprocessed images separately, and performs guided filtering processing on each group of preprocessed images using the guide maps. There is no data dependency between groups, allowing for simultaneous multi-threaded computation and saving preprocessing time in the algorithm.

Description

Technical Field

[0001] This invention relates to the field of high-resolution image reconstruction technology, specifically to an image denoising method and an imaging method. Background Technology

[0002] Fourier layered microscopy (FPM) is a novel computational imaging technique developed in recent years. It uses iterative algorithms to gradually recover intensity and phase distribution, achieving large field-of-view and high-resolution imaging capabilities. Although FPM was only proposed in 2013, it has already been successfully applied in the field of microscopy. Its idea of ​​improving spatial resolution is similar to synthetic aperture technology, making it a promising new high-resolution optical imaging technique based on phase information for far-field applications. However, with increasing imaging distance and complex imaging environments, the reconstruction results are largely limited by the image quality of the original acquired images, necessitating image processing for improvement. For imaging processes like Fourier layered imaging, which utilize a series of low-resolution images for reconstruction, conventional laser active imaging denoising methods are difficult to apply. Therefore, it is necessary to research specialized denoising algorithms for far-field Fourier layered imaging.

[0003] Currently, there is very little research on the application of Fourier layered imaging methods in the far field. In 2015, Bian et al. incorporated Gaussian noise into the evaluation function and proposed an FPM reconstruction method based on the Wirtinger flow algorithm using gradient descent for updating. This method can handle Gaussian noise well, but it requires precise initialization and a long number of iterations. Subsequently, the same researchers proposed an FPM reconstruction algorithm based on Poisson maximum likelihood estimation for modeling and using truncated Wirtinger for updating, which can handle Poisson and speckle noise well. This algorithm also requires many iterations to converge. In 2017, Zhang et al. considered both Gaussian and Poisson noise and proposed an FPM reconstruction algorithm based on generalized Anscombe transform using maximum likelihood theory. This algorithm also requires many iterations to converge. These algorithms are all noise reduction methods for Fourier layered imaging systems based on LED array illumination, and no research has been conducted on reflective long-range Fourier layered imaging systems based on laser illumination. Summary of the Invention

[0004] This application provides an image denoising method that can solve the technical problem of poor denoising effect.

[0005] This application also provides an imaging method that can solve the technical problems of low reconstructed image resolution and long algorithm running time.

[0006] On one hand, embodiments of this application provide an image denoising method, including:

[0007] Acquire multiple preprocessed images.

[0008] The preprocessed images are divided into multiple groups, with each group containing multiple preprocessed images.

[0009] For each group of preprocessed images, a guide image is obtained based on all the preprocessed images in the group, resulting in multiple guide images. These multiple guide images correspond one-to-one with multiple groups of preprocessed images.

[0010] Guided filtering is performed on each preprocessed image within a group using a guide image corresponding to each group of preprocessed images.

[0011] According to an embodiment of the first aspect of this application, a guide image is obtained based on all preprocessed images within the group, including:

[0012] The average pixel value of all preprocessed images in the group is calculated based on pixel coordinates to obtain the mean image.

[0013] The mean image is transformed to grayscale to obtain the guide image corresponding to the preprocessed image set.

[0014] According to any of the foregoing embodiments of the first aspect of this application, guided filtering is performed on each preprocessed image within a group using a guide map corresponding to each group of preprocessed images, including:

[0015] The guidance coefficients are calculated by combining the preprocessed image and the guidance map corresponding to the group in which the preprocessed image belongs. Each preprocessed image corresponds to a set of guidance coefficients.

[0016] The pixel values ​​of the corresponding preprocessed image are linearly transformed using the guiding coefficient to obtain the filtered preprocessed image.

[0017] Secondly, embodiments of this application provide an imaging method, including...

[0018] Acquire multiple raw images.

[0019] Regions of interest (ROIs) are extracted from multiple original images to obtain multiple preprocessed images.

[0020] One of the multiple preprocessed images is selected as the reference image, and the remaining preprocessed images are registered.

[0021] The registered preprocessed images are divided into multiple groups, each group containing multiple preprocessed images.

[0022] For each group of preprocessed images, a guide image is obtained based on all the preprocessed images in the group, resulting in multiple guide images. These multiple guide images correspond one-to-one with multiple groups of preprocessed images.

[0023] Guided filtering is performed on each preprocessed image within a group using a guide image corresponding to each group of preprocessed images.

[0024] The filtered preprocessed image is analyzed and processed using the Fourier layer imaging method to obtain a clear image.

[0025] According to an embodiment of the second aspect of this application, a guide image is obtained based on all preprocessed images within the group, including:

[0026] The average pixel value of all preprocessed images in the group is calculated based on pixel coordinates to obtain the mean image.

[0027] The mean image is transformed to grayscale to obtain the guide image corresponding to the preprocessed image set.

[0028] According to any of the foregoing embodiments of the second aspect of this application, guided filtering is performed on each preprocessed image within a group using a guide map corresponding to each group of preprocessed images, including:

[0029] The guidance coefficients are calculated by combining the preprocessed image and the guidance map corresponding to the group in which the preprocessed image belongs. Each preprocessed image corresponds to a set of guidance coefficients.

[0030] The pixel values ​​of the corresponding preprocessed image are linearly transformed using the guiding coefficient to obtain the filtered preprocessed image.

[0031] According to any of the foregoing embodiments of the second aspect of this application, the multiple original images are a sequence of original images obtained by the imaging device performing moving scans according to a fixed step size, and the overlap rate between two adjacent original images is... .

[0032] According to any of the foregoing embodiments of the second aspect of this application, the size of the guided filtering window is based on the overlap rate between the two original images. The pixel dimensions of the preprocessed image are determined.

[0033] According to any of the foregoing embodiments of the second aspect of this application, the formula for calculating the size of the filter window for guided filtering is as follows:

[0034]

[0035] in, The radius of the guided filter window is represented by M, and M and N represent the pixel dimensions of the preprocessed image. Represents an integer not exceeding x.

[0036] According to any of the foregoing embodiments of the second aspect of this application, before analyzing and processing the filtered preprocessed image using the Fourier layered imaging method to obtain a clear image, the method further includes...

[0037] Homomorphic filtering is applied to each group of preprocessed images, or

[0038] Homomorphic filtering is performed on all preprocessed images simultaneously.

[0039] According to any of the foregoing embodiments of the second aspect of this application, the step of selecting one of the plurality of preprocessed images as a reference image and registering the remaining preprocessed images includes...

[0040] Select a preprocessed image as the reference image, and then crop the target region from the reference image to obtain a new preprocessed image.

[0041] The remaining preprocessed images to be registered are processed as follows:

[0042] Calculate the cross-power spectrum between the reference image and the preprocessed image to be registered.

[0043] An approximate two-dimensional impulse function is obtained by performing an inverse Fourier transform on the cross power spectrum.

[0044] The peak value of the approximate two-dimensional impulse function is calculated to obtain the displacement of the preprocessed image to be registered.

[0045] The target region is cropped from the preprocessed image to be registered based on the displacement, resulting in a new preprocessed image.

[0046] According to any of the foregoing embodiments of the second aspect of this application, after selecting one of the plurality of preprocessed images as a reference image and registering the remaining preprocessed images, the method further includes...

[0047] Global noise reduction is performed on the new preprocessed image obtained after registration.

[0048] The image denoising method of this application performs guided filtering on preprocessed images in groups, calculates the guided image for each group of preprocessed images, and performs guided filtering on each group of preprocessed images through the guided image. There is no data dependency between groups, and multi-threaded calculation can be performed simultaneously, saving the preprocessing time of the algorithm.

[0049] The imaging method of this application uses the denoised sequence of images as input to the Fourier stacked imaging method. Compared with the input of images without denoising processing, it can effectively reduce the number of algorithm iterations and improve the image restoration quality. Attached Figure Description

[0050] Figure 1a For the original image, Figure 1b It is a preprocessed image after noise reduction through guided filtering and homomorphic filtering;

[0051] Figure 2aThe original image sequence, Figure 2b It is a preprocessed image sequence after noise reduction through guided filtering and homomorphic filtering;

[0052] Figure 3a To directly utilize the results of the Fourier layered imaging method iteration, Figure 3b It is the result of iterative Fourier layered imaging after first undergoing guided filtering, then homomorphic filtering for noise reduction;

[0053] Figure 4a This is the result of iterative Fourier layered imaging after only performing guided filtering. Figure 4b It is the result of iterative Fourier layered imaging after first undergoing guided filtering, then homomorphic filtering for noise reduction;

[0054] Figure 5a It is the result of iterative processing using the Fourier layered imaging method, after first applying homomorphic filtering and then guided filtering. Figure 5b It is the result of iterative Fourier layer imaging after first undergoing guided filtering, then homomorphic filtering for noise reduction. Detailed Implementation

[0055] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0056] The first aspect of this application provides an image noise reduction method, including the following steps:

[0057] S1. Obtain multiple preprocessed images; the preprocessed images can be images that have undergone certain preprocessing, such as ROI extraction, registration, etc.

[0058] S2. Divide multiple preprocessed images into multiple groups, each group including multiple preprocessed images; there can be multiple grouping methods, for example, the number of preprocessed images contained in each group can be the same or different; preferably, each group contains the same number of preprocessed images; the order of all preprocessed images can be the order in which the camera device acquires the original images.

[0059] Suppose that m*n (m rows, n columns) preprocessed images are divided into k*l groups (k rows, l columns), and the number of preprocessed images contained in each group is:

[0060] (1)

[0061] (2)

[0062] in, Represents an integer not exceeding x.

[0063] S3. For each group of preprocessed images, obtain a guide image based on all the preprocessed images in the group, resulting in multiple guide images. These multiple guide images correspond one-to-one with the multiple groups of preprocessed images. In this step, a guide image can be calculated based on each group of preprocessed images.

[0064] S4. Guided filtering is performed on each preprocessed image in the group using the guide image corresponding to each group of preprocessed images.

[0065] The image denoising method of the first aspect of this application performs guided filtering on preprocessed images in groups, calculates a guided image for each group of preprocessed images, and performs guided filtering on each group of preprocessed images through the guided image. There is no data dependency between groups, and multi-threaded calculation can be performed simultaneously, saving the preprocessing time of the algorithm.

[0066] In some embodiments, step S3, "obtaining a guide image based on all preprocessed images within the group," includes:

[0067] S31. Calculate the average pixel value of all preprocessed images in the group according to pixel coordinates to obtain the mean image. Assuming that each preprocessed image contains M rows of pixels, and each row contains N pixels, then each preprocessed image contains M*N pixels, where M and N represent the pixel dimensions of the preprocessed image. The mean image also contains M*N pixels, and the pixel value of each pixel is equal to the sum of the pixel values ​​of the corresponding pixels in each preprocessed image and then the average.

[0068] S32. Perform a grayscale transformation on the mean image to obtain the guide image corresponding to the preprocessed image group. The grayscale transformation can be a power transform, logarithmic transform, or gamma transform. Preferably, a gamma transform is used to perform the grayscale transformation on the mean image; then the formula for calculating the guide image of the preprocessed image group in the i-th row and j-th column is:

[0069] (3)

[0070] The guide map calculated using the above method has a good filtering effect. To improve the quality of the preprocessed image, the guide map can effectively reduce speckle noise and random noise generated by laser active imaging, as well as enhance image details.

[0071] Of course, there are other methods to calculate the guide map. For example, the method of calculating the average pixel value in step P31 can be replaced by calculating the weighted mean, calculating the median, etc.

[0072] In some embodiments, step S4 includes the following steps:

[0073] S41. Calculate the guiding coefficients by combining the preprocessed image and the guiding image corresponding to the group containing the preprocessed image. Each preprocessed image corresponds to a set of guiding coefficients.

[0074] S42. Apply a linear transformation to the pixel values ​​of the corresponding preprocessed image using the guiding coefficient to obtain the filtered preprocessed image. The formula for guiding filtering for each preprocessed image in the i-th row and j-th column group is as follows:

[0075] (4)

[0076] in, and The guiding coefficient is calculated using the following formula:

[0077] (5)

[0078] (6)

[0079] in, This represents the variance of the image obtained after multiplying the preprocessed image and the guide image within the corresponding filter window. This represents the variance of the guide plot within the corresponding filter window. Represents the regularization factor. This represents the mean of the guide plot within the corresponding filter window. This represents the mean of the preprocessed image within the corresponding filter window.

[0080] A second aspect of this application provides an imaging method, including the steps of:

[0081] P1. Acquire multiple original images; these original images are a sequence of images obtained by the camera device moving and scanning at a fixed step size, with an overlap rate between adjacent original images of [missing information]. For example, the camera device can move along a serpentine path with a fixed step size, or it can move along a zigzag path with a fixed compensation. The area captured after each step overlaps with the area captured in the previous step, thus creating an overlap between the two original images captured, in order to achieve spectral splicing in Fourier space.

[0082] P2. Extract the ROI regions from multiple original images to obtain multiple preprocessed images; the target surface is located within the area captured by the camera, and the extracted ROI region is the local area where the target surface is located.

[0083] P3. Select one of the multiple preprocessed images as the reference image and register the remaining preprocessed images. Since positional deviations may occur during the movement of the camera device, registration can eliminate the deviations caused by the movement of the camera device.

[0084] P4. Divide the registered preprocessed images into multiple groups, each group containing multiple preprocessed images; the grouping method is as described above and will not be repeated here.

[0085] P5. For each group of preprocessed images, obtain a guide image based on all the preprocessed images in the group to obtain multiple guide images. The multiple guide images correspond one-to-one with the multiple groups of preprocessed images.

[0086] P6. Use the guide image corresponding to each group of preprocessed images to perform guided filtering on each preprocessed image in the group;

[0087] P7. The filtered preprocessed image is analyzed and processed using the Fourier layered imaging method to obtain a clear image.

[0088] The imaging method of this application uses the sequence of images after guided filtering and noise reduction as the input of the Fourier stacked imaging method. Compared with the input of images without noise reduction processing, it can effectively reduce the number of algorithm iterations and improve the image restoration quality.

[0089] In some embodiments, step P5, "obtaining a guide image based on all preprocessed images within the group," includes:

[0090] P51. Calculate the average pixel value of all preprocessed images in the group according to pixel coordinates to obtain the mean image.

[0091] P52. Perform grayscale transformation on the mean image to obtain the guide image corresponding to the preprocessed image set.

[0092] The specific implementation methods and beneficial effects of steps P51 and P52 are the same as those of steps S31 and S32, and will not be repeated here.

[0093] In some embodiments, step P6 includes the following steps:

[0094] P61. Calculate the guiding coefficients by combining the preprocessed image and the guiding image corresponding to the group containing the preprocessed image. Each preprocessed image corresponds to a set of guiding coefficients.

[0095] P62. The pixel values ​​of the corresponding preprocessed image are linearly transformed using the guiding coefficient to obtain the filtered preprocessed image.

[0096] The specific implementation methods and beneficial effects of steps P61 and P62 are the same as those of steps S41 and S42, and will not be repeated here.

[0097] In some embodiments, the size of the guided filter window is based on the overlap rate between the two original images. The pixel dimensions of the preprocessed image are determined. Specifically, the formula for calculating the size of the guided filter window is as follows:

[0098] (7)

[0099] in, The radius of the guided filter window is represented by M, and M and N represent the pixel dimensions of the preprocessed image. This represents an integer not exceeding x. In this embodiment, the size of the filtering window can be adaptively adjusted according to the image acquisition parameters.

[0100] In some embodiments, the following step is included before step P7:

[0101] P81. Perform homomorphic filtering on each group of preprocessed images separately; performing homomorphic filtering separately within each group can improve filtering efficiency.

[0102] or

[0103] P82. Perform homomorphic filtering on all preprocessed images simultaneously; the same filtering effect can be achieved without using grouping for homomorphic filtering, but the preprocessing time will be longer.

[0104] Homomorphic filtering employs a high-pass filter, such as a Butterworth high-pass filter or a trapezoidal high-pass filter. Homomorphic filtering can further remove noise while simultaneously compressing the dynamic range and enhancing contrast.

[0105] Homomorphic filtering can be performed before or after guided filtering. Preferably, homomorphic filtering is performed after guided filtering. Using this embodiment, the high-resolution image recovered through guided filtering followed by homomorphic filtering exhibits significantly reduced noise and a marked improvement in image resolution.

[0106] In some embodiments, step P3 includes the following steps:

[0107] P31. Select a preprocessed image as the reference image, and crop the target region from the reference image to obtain a new preprocessed image.

[0108] The remaining preprocessed images to be registered are processed as follows:

[0109] P32. Calculate the cross-power spectrum between the reference image and the preprocessed image to be registered; the specific calculation formula is as follows:

[0110] (8)

[0111] in, yes conjugate, It is the i-th image f i The Fourier transform of (x,y), where M and N are the image dimensions, x0 is the horizontal displacement, and y0 is the vertical displacement.

[0112] P33. An inverse Fourier transform of the cross power spectrum yields an approximate two-dimensional impulse function; the specific formula is as follows:

[0113] (9)

[0114] P34. Solve for the peak value of the approximate two-dimensional impulse function to obtain the displacement of the preprocessed image to be registered;

[0115] P35. Based on the displacement, the target region is cropped from the preprocessed image to be registered to obtain a new preprocessed image;

[0116] Using the top left corner of the preprocessed image to be registered as the origin, the target region is cropped from the point obtained by shifting x0 to the left and y0 down, thus obtaining a new preprocessed image.

[0117] In some embodiments, after step P3, the following step is further included:

[0118] P9. Perform global noise reduction on the new preprocessed image obtained after registration. Global noise reduction involves subtracting the corresponding black level. The black level value is calculated during the camera device calibration process, i.e., by capturing several images without lighting and statistically analyzing the average pixel value to obtain the black level magnitude.

[0119] The imaging method in this application employs a global denoising + guided filtering + homomorphic filtering approach, effectively reducing noise in each laser image and making the image clearer. This provides technical support for applications such as far-field Fourier layered imaging, remote sensing imaging, and detection of faint targets in space. The denoised image sequence, used as input to the Fourier layered imaging iterative algorithm, effectively reduces the number of iterations and improves image restoration quality compared to input images without denoising processing.

[0120] In this application, the original image can be an image of a diffuse reflection target obtained by a meter-level active illumination long-range imaging system. This means the distance between the target and the illumination system is greater than 1 meter, and the target surface is non-reflective, resulting in diffuse reflection of the illumination light. Due to environmental interference and limitations in the imaging hardware system during long-range imaging, the original image is a superposition of real image information and various noise sources, leading to a significant degraded image quality. Therefore, noise reduction processing is needed to improve image quality. The mixed noise includes speckle noise from the laser, thermal noise from the image sensor, and quantization noise during image digitization. In this application, guided filtering is primarily used to reduce speckle noise from the laser, global noise reduction is used to reduce thermal noise from the image sensor, and homomorphic filtering is used to reduce quantization noise during image digitization. After global noise reduction, guided filtering, and homomorphic filtering, the noise of the preprocessed image is significantly reduced.

[0121] like Figure 1a and Figure 1b As shown, Figure 1a The original captured image. Figure 1b The image is after noise reduction through guided filtering and homomorphic filtering. As can be seen from the image, the noise is effectively suppressed and the image clarity is significantly increased after guided filtering and homomorphic filtering.

[0122] like Figure 2a and Figure 2b As shown, Figure 2a This is the original acquired image sequence. Figure 2b This is an image sequence after noise reduction through guided filtering and homomorphic filtering; as can be seen from the figure, the noise in each image is effectively suppressed after guided filtering and homomorphic filtering.

[0123] like Figure 3a and Figure 3b As shown, Figure 3a To directly utilize the results of the Fourier layered imaging method iteration, Figure 3b The image is the result of iterative Fourier layer imaging after first undergoing guided filtering and then homomorphic filtering for noise reduction. As can be seen from the image, the image at the iteration point of the Fourier layer imaging method is very clear after guided filtering and homomorphic filtering.

[0124] like Figure 4a and Figure 4b As shown, Figure 4a This is the result of iterative Fourier layered imaging after only performing guided filtering. Figure 4b It is the result of iterative Fourier layered imaging after first undergoing guided filtering, then homomorphic filtering for noise reduction; as can be seen from the figure. Figure 4b The clarity is significantly higher than Figure 4aThis indicates that the filtering effect of guided filtering + homomorphic filtering is very significant.

[0125] like Figure 5a and Figure 5b As shown, Figure 5a It is the result of iterative processing using the Fourier layered imaging method, after first applying homomorphic filtering and then guided filtering. Figure 5b The image is the result of iterative Fourier layer imaging after first applying guided filtering and then homomorphic filtering for noise reduction. As can be seen from the figure, changing the order of homomorphic filtering and guided filtering results in different levels of image sharpness; the noise reduction effect of applying guided filtering first and then homomorphic filtering is better than applying homomorphic filtering first and then guided filtering.

Claims

1. An imaging method, characterized in that, include Acquire multiple raw images. Regions of interest (ROIs) are extracted from multiple original images to obtain multiple preprocessed images. One of the preprocessed images is selected as the reference image, and the remaining preprocessed images are registered. The registered preprocessed images are divided into multiple groups, each group containing multiple preprocessed images. For each group of preprocessed images, a guide image is obtained based on all the preprocessed images in the group, resulting in multiple guide images. These multiple guide images correspond one-to-one with multiple groups of preprocessed images. Guided filtering is performed on each preprocessed image within a group using a guide image corresponding to each group of preprocessed images. The formula for calculating the size of the filter window in the guided filter is as follows: ； in, The radius of the guided filter window is represented by M, and M and N represent the pixel dimensions of the preprocessed image. Represents an integer not exceeding x; α is the overlap rate between two adjacent original images; Homomorphic filtering is applied to each group of preprocessed images separately, or homomorphic filtering is applied to all preprocessed images simultaneously. The filtered preprocessed image is analyzed and processed using the Fourier layer imaging method to obtain a clear image.

2. The imaging method according to claim 1, characterized in that, The step of obtaining a guide image based on all preprocessed images within the group includes: The average pixel value of all preprocessed images in the group is calculated based on pixel coordinates to obtain the mean image. The mean image is transformed to grayscale to obtain the guide image corresponding to the preprocessed image set.

3. The imaging method according to claim 2, characterized in that, The step of using a guide map corresponding to each group of preprocessed images to perform guided filtering on each preprocessed image within a group includes: The guidance coefficients are calculated by combining the preprocessed image and the guidance map corresponding to the group in which the preprocessed image belongs. Each preprocessed image corresponds to a set of guidance coefficients. The pixel values ​​of the corresponding preprocessed image are linearly transformed using the guiding coefficient to obtain the filtered preprocessed image.

4. The imaging method according to claim 1, characterized in that, The multiple original images are a sequence of original images obtained by the camera device moving and scanning at a fixed step size.

5. The imaging method according to claim 1, characterized in that, The step of selecting one of the multiple preprocessed images as a reference image and registering the remaining preprocessed images includes... Select a preprocessed image as the reference image, and then crop the target region from the reference image to obtain a new preprocessed image. The remaining preprocessed images to be registered are processed as follows: Calculate the cross-power spectrum between the reference image and the preprocessed image to be registered. An approximate two-dimensional impulse function is obtained by performing an inverse Fourier transform on the cross power spectrum. The peak value of the approximate two-dimensional impulse function is calculated to obtain the displacement of the preprocessed image to be registered. The target region is cropped from the preprocessed image to be registered based on the displacement, resulting in a new preprocessed image.

6. The imaging method according to claim 1, characterized in that, After selecting one of the multiple preprocessed images as the reference image and registering the remaining preprocessed images, the method further includes global noise reduction on the newly preprocessed image obtained after registration.

Images