A brain tumor tissue hyperspectral data non-uniform noise calibration method

Abstract

The present application relates to noise calibration method, specifically to a kind of brain tumor tissue hyperspectral data non-uniform noise calibration method, solve the uniformity output response of no light source and the calibration method based on mathematical statistics analysis, when carrying out brain tumor tissue non-uniformity stripe noise elimination, there is the problem of not ideal elimination effect.The present application first collects sample image and corresponding dark background image in the same environment, then dark background image is weighted least square filtering processing, the dark background image after filtering processing can keep edge information, then the original dark background image is subtracted from the image after filtering processing to obtain edge information, the weighted coefficient of corresponding pixel position is obtained by taking mean value to all wave bands of edge information, then the fixed noise content of instrument is separated wave by wave according to the weighted coefficient;For sample image, it is weighted least square filtering processing of embedded bilateral filtering, then the final clear image is obtained by subtracting the fixed noise content from the sample image after filtering processing.

Description

Technical Field

[0001] This invention relates to noise calibration methods, specifically to a method for calibrating non-uniform noise in hyperspectral data of brain tumor tissue. Background Technology

[0002] Hyperspectral imaging, due to its ability to display rich spatial and spectral information, is commonly used in imaging cancerous tissues. However, in daily use, hyperspectral imaging system sensors exhibit non-uniform response outputs due to the influence of multiple factors such as ambient temperature. This non-uniformity reduces the imaging quality and spectral accuracy of hyperspectral images, severely hindering information extraction and target recognition in cancerous samples, especially brain tumor tissue. Therefore, effective non-uniformity correction is essential in hyperspectral imaging systems to improve the accuracy of quantitative image analysis. Currently, in commonly used spatial pushbroom hyperspectral cancerous tissue imaging systems, image non-uniformity manifests as stripe noise throughout the sampled image. Therefore, removing stripe noise can largely achieve non-uniformity correction in hyperspectral images.

[0003] There are two main types of methods for removing non-uniform stripe noise from brain tumor tissue sample images. One type uses a uniform output response without a light source as a reference image to eliminate non-uniform stripe noise. The problem with this method is that it doesn't estimate the stripe noise in the dark background based on the location of dark pixels, resulting in unsatisfactory noise removal. The other type uses mathematical statistical analysis to process the sample image, mainly through filtering and transformation. However, this method doesn't consider the characteristics of the instrument itself, thus its effectiveness in eliminating non-uniform noise is minimal.

[0004] Therefore, in order to better achieve non-uniform noise removal, the non-uniform noise removal method of the spatial pushbroom hyperspectral imaging system must combine the two methods mentioned above. On the one hand, it should focus on the instrument imaging mechanism and analyze the causes of non-uniform noise. On the other hand, it should analyze the data patterns from a mathematical and statistical perspective, so as to organically combine the front-end imaging and the back-end data processing, thereby improving the non-uniform noise elimination effect. Summary of the Invention

[0005] The purpose of this invention is to address the problem that both uniform output response without light source and calibration methods based on mathematical statistical analysis have unsatisfactory elimination effects when eliminating non-uniform stripe noise in brain tumor tissue, and to provide a calibration method for non-uniform noise in hyperspectral data of brain tumor tissue.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0007] A method for calibrating non-uniform noise in hyperspectral data of brain tumor tissue, characterized by the following steps:

[0008] S1: Collect sample images and corresponding dark background images under the same environment;

[0009] S2: Perform weighted least squares filtering on the dark background image;

[0010] S3: Subtract the original dark background image from the dark background image processed by weighted least squares filtering to obtain the edge information of the dark background image, and take the average value of the edge information across the entire band to obtain the weighting coefficient of the corresponding pixel position.

[0011] S4: Based on the weighting coefficients of the corresponding pixel positions, separate the fixed noise content of the instrument band by band;

[0012] S5: Perform weighted least squares filtering with embedded bilateral filtering on the sample image;

[0013] S6: Subtract the filtered sample image from the fixed noise content of the separated instrument to obtain the final clear image, thus completing the non-uniform noise calibration of the hyperspectral data of brain tumor tissue.

[0014] Further, step S1 specifically involves: acquiring sample images χ under the same environment. (i,j,B) and the corresponding dark background image x (i,j,B) , where (i,j) is the spatial position and B is the spectral position.

[0015] Further, step S2 specifically involves performing a weighted least squares filter on the dark background image to obtain the filtered dark background image y. (i,j,B) ;

[0016]

[0017] in: This represents the gradient of the filtered dark background image along the X direction. The gradient along the Y direction of the filtered dark background image, * represents the pixel value, and ω i ω represents the weight of the dark background image in the X direction. j This represents the weight of the dark background image in the Y direction. and ε is the universal weight, λ is a small coefficient, λ is the regularization coefficient, and y is the dark background image after filtering.

[0018] Further, step S3 specifically involves: processing the original dark background image x (i,j,B) Compared with the dark background image y after weighted least squares filtering (i,j,B)Subtraction yields the edge information e′ of the dark background image. (i,j,B) and edge information e′ (i,j,B) The average value across the entire band is used to obtain the weighting coefficient K for the corresponding pixel location. (i,j,B) ;

[0019] e′ (i,j,B) =x (i,j,B) -y (i,j,B)

[0020]

[0021] Where: M×N is the size of the dark background image.

[0022] Further, step S4 specifically involves: based on the obtained weighting coefficient K for the corresponding pixel position... (i,j,B) The fixed noise content N of the instrument is separated band by band. (i,j,B) ;

[0023] N (i,j,B) =K (i,j,B) ×x (i,j,B) / (dark_m) B

[0024] Where: (dark_m) B This represents the mean value of a single band in a dark background image.

[0025] Further, step S5 specifically involves: performing weighted least squares filtering with embedded bilateral filtering on the sample image to obtain the filtered sample image γ. (i,j,B) ;

[0026]

[0027] Where: χ (i,j) For the sample image, γ (i,j) The sample image after filtering. For the gradient of the sample image, Let ω be the gradient of the filtered sample image. x,(i,j) ω represents the weight of the filtered sample image along the X direction. y,(i,j) f represents the weights along the Y-direction of the filtered sample image. BLF This represents the weights of the two-sided operators.

[0028] Further, step S6 specifically involves: processing the filtered sample image γ (i,j,B) The fixed noise content N of the separated instrument (i,j,B) Subtracting them yields the final, clear image X′. (i,j,B) ;

[0029] X′ (i,j,B) =γ(i,j,B) -N (i,j,B)

[0030] Complete the non-uniform noise calibration of hyperspectral data of brain tumor tissue.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] (1) The present invention provides a method for non-uniform noise calibration of hyperspectral data of brain tumor tissue. First, sample images and dark background images of brain tumor tissue are collected under the same environment. This ensures that the influencing factors of the two images are basically the same, which makes the calculation more accurate in the later stage. After the sample image and dark background image are collected, the two images are filtered respectively. Weighted least squares filtering is used for the dark background image. This filtering method has the characteristic of preserving edge information. This filtering method can preserve the edge information of the original dark background image to the maximum extent. Then, the edge information is obtained by subtracting the filtered dark background image from the original dark background image. The edge information is then averaged across the entire band to obtain the weighting coefficient of the corresponding pixel position. The fixed noise content of the instrument is then separated band by band according to the obtained weighting coefficient. At this time, the processing calculation of the dark background image is completed.

[0033] (2) The present invention provides a non-uniform noise calibration method for hyperspectral data of brain tumor tissue. For the sample image, a weighted least squares filter with embedded bilateral filtering is used. After obtaining the filtered sample image, it is subtracted from the fixed noise content of the instrument separated in the dark background image to obtain a clear image.

[0034] (3) The present invention provides a non-uniform noise calibration method for hyperspectral data of brain tumor tissue. Based on the imaging characteristics of the spatial dimension push-broom hyperspectral brain tumor tissue imaging system, it can more accurately separate the fixed noise content of the instrument itself from the sample image, thereby obtaining a clean and clear hyperspectral brain tumor image. Attached Figure Description

[0035] Figure 1 This is a schematic diagram illustrating the working principle of an embodiment of a non-uniform noise calibration method for hyperspectral data of brain tumor tissue according to the present invention.

[0036] Figure 2 This is a comparison of the grayscale distribution of brain tumor tissue before and after denoising processing using an embodiment of the non-uniform noise calibration method for hyperspectral data of brain tumor tissue according to the present invention. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and exemplary embodiments.

[0038] The working principle diagram of the non-uniform noise calibration method for hyperspectral data of brain tumor tissue of the present invention is shown below. Figure 1 As shown, the first step is to collect sample images of brain tumor tissue and dark background images under the same environment to ensure that the influencing factors are basically the same. After collecting the sample images and dark background images, the two images are filtered separately.

[0039] The dark background image is processed using weighted least squares filtering. Based on the characteristic that weighted least squares filtering preserves edge information, this filtering method can retain the edge information of the original dark background image to the maximum extent. Then, the filtered dark background image is subtracted from the original dark background image to obtain the edge information. The average value of the obtained edge information across the entire band is then taken to obtain the weighting coefficient of the corresponding pixel position. Based on the obtained weighting coefficient, the fixed noise content of the instrument is separated band by band. At this point, the processing calculation of the dark background image is completed.

[0040] For the sample image, a weighted least squares filter with embedded bilateral filtering is used. After obtaining the filtered sample image, it is subtracted from the fixed noise content of the instrument separated from the dark background image to obtain a clear image.

[0041] Specifically, the following steps are included:

[0042] S1: Sample images were acquired under the same conditions. (i,j,B) and the corresponding dark background image x (i,j,B) , where (i,j) is the spatial position and B is the spectral position.

[0043] S2: Perform weighted least squares filtering on the dark background image to obtain the filtered dark background image y. (i,j,B) ;

[0044]

[0045] in: This represents the gradient of the filtered dark background image along the X direction. The gradient along the Y direction of the filtered dark background image, * represents the pixel value, and ω i ω represents the weight of the dark background image in the X direction. j This represents the weight of the dark background image in the Y direction. and ε is the universal weight, λ is a small coefficient, λ is the regularization coefficient, and y is the dark background image after filtering.

[0046] S3: Extract the original dark background image x (i,j,B) Compared with the dark background image y after weighted least squares filtering (i,j,B)Subtraction yields the edge information e′ of the dark background image. (i,j,B) and edge information e′ (i,j,B) The average value across the entire band is used to obtain the weighting coefficient K for the corresponding pixel location. (i,j,B) ;

[0047] e′ (i,j,B) =x (i,j,B) -y (i,j,B)

[0048]

[0049] Where: M×N is the size of the dark background image.

[0050] S4: Based on the weighting coefficient K obtained for the corresponding pixel position (i,j,B) The fixed noise content N of the instrument is separated band by band. (i,j,B) ;

[0051] N (i,j,B) =K (i,j,B) ×x (i,j,B) / (dark_m) B

[0052] Where: (dark_m) B This represents the mean value of a single band in a dark background image.

[0053] S5: Perform weighted least squares filtering with embedded bilateral filtering on the sample image to obtain the filtered sample image γ. (i,j,B) ;

[0054]

[0055] Where: χ (i,j) For the sample image, γ (i,j) The sample image after filtering. For the gradient of the sample image, Let ω be the gradient of the filtered sample image. x,(i,j) ω represents the weight of the filtered sample image along the X direction. y,(i,j) f represents the weights along the Y-direction of the filtered sample image. BLF This represents the weights of the two-sided operators.

[0056] S6: Filter the sample image γ (i,j,B) The fixed noise content N of the separated instrument (i,j,B) Subtraction yields the final, clear image X. ′ (i,j,B) ;

[0057] X ′ (i,j,B) =γ(i,j,B) -N (i,j,B)

[0058] Complete the non-uniform noise calibration of hyperspectral data of brain tumor tissue.

[0059] This yields a clear image X. ′ (i,j,B) The grayscale distribution comparison of brain tumor tissue before and after denoising is shown in the figure below. Figure 2 As shown, it is clear that the image after noise removal is much clearer.

[0060] In step S5, the sample image undergoes weighted least squares filtering with embedded bilateral filtering. The theoretical calculation method for the formula is as follows:

[0061] The matrix representation of the filtered sample image is as follows:

[0062] (χ-γ) T (χ-γ)+λ(χ T D x T A x D x χ+f BLF (γ T D y T A y D y γ))

[0063] Among them: A x A y Each of them is a k×k diagonal matrix, D x D y Matrix representations of discrete difference operators

[0064] The dark background image is M×N in size. We introduce a variable p = M×N, where p is a one-dimensional vector. Setting the gradient of the filtered sample image to 0, we obtain the equation:

[0065] (I+λf BLF (L γ ))χ=γ

[0066] In the formula L γ =D T A x D x +D T A y D y χ represents the sample image, and γ represents the filtered sample image. Regarding the gradient design of the weights, this invention does not use the logarithmic brightness of the WLS algorithm, but instead uses the gradient of the original image. Verification has shown that this achieves a smoothing effect more quickly. The weight design is as follows:

[0067]

[0068] Where: α x,(i,j) To correct the lateral gradient, α y,(i,j) To correct the longitudinal gradient, For the original lateral gradient, ε is the original longitudinal gradient, and ε is a small coefficient.

[0069] The embodiments described above are merely illustrative of specific implementations of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for calibrating non-uniform noise in hyperspectral data of brain tumor tissue, characterized in that, Includes the following steps: S1: Collect sample images and corresponding dark background images under the same environment; Specifically, this involves acquiring sample images χ under the same environment. (i,j,B) and the corresponding dark background image x (i,j,B) , where (i,j) is the spatial position and B is the spectral position; S2: Perform weighted least squares filtering on the dark background image; Specifically, a weighted least squares filter is applied to the dark background image to obtain the filtered dark background image y. (i,j,B) ; in: This represents the gradient of the filtered dark background image along the X direction. The gradient along the Y direction of the filtered dark background image, * represents the pixel value, and ω i ω represents the weight of the dark background image in the X direction. j This represents the weight of the dark background image in the Y direction. and ε is the universal weight, λ is a small coefficient, λ is the regularization coefficient, and y is the dark background image after filtering. S3: Subtract the original dark background image from the dark background image processed by weighted least squares filtering to obtain the edge information of the dark background image, and take the average value of the edge information across the entire band to obtain the weighting coefficient of the corresponding pixel position. Specifically, it involves: transforming the original dark background image x (i,j,B) Compared with the dark background image y after weighted least squares filtering (i,j,B) Subtraction yields the edge information e′ of the dark background image. (i,j,B) and edge information e′ (i,j,B) The average value across the entire band is used to obtain the weighting coefficient K for the corresponding pixel location. (i,j,B) ; and' (i,j,B) =x (i,j,B) -and (i,j,B) Where: M×N is the size of the dark background image; S4: Based on the weighting coefficients of the corresponding pixel positions, separate the fixed noise content of the instrument band by band; S5: Perform weighted least squares filtering with embedded bilateral filtering on the sample image; S6: Subtract the filtered sample image from the fixed noise content of the separated instrument to obtain the final clear image, thus completing the non-uniform noise calibration of the hyperspectral data of brain tumor tissue.

2. The method for calibrating non-uniform noise in hyperspectral data of brain tumor tissue according to claim 1, characterized in that, Step S4 is as follows: Based on the weighting coefficient K obtained for the corresponding pixel position (i,j,B) The fixed noise content N of the instrument is separated band by band. (i,j,B) ; N (i,j,B) =K (i,j,B) ×x (i,j,B) / (dark_m) B Where: (dark_m) B This represents the mean value of a single band in a dark background image.

3. The method for calibrating non-uniform noise in hyperspectral data of brain tumor tissue according to claim 2, characterized in that, Step S5 is as follows: The sample image is subjected to weighted least squares filtering with embedded bilateral filtering to obtain the filtered sample image γ. (i,j,B) ; Where: χ (i,j) For the sample image, γ (i,j) The sample image after filtering. For the gradient of the sample image, Let ω be the gradient of the filtered sample image. x,(i,j) ω represents the weight of the filtered sample image along the X direction. y,(i,j) f represents the weights along the Y-direction of the filtered sample image. BLF This represents the weights of the two-sided operators.

4. The method for calibrating non-uniform noise in hyperspectral data of brain tumor tissue according to claim 3, characterized in that, Step S6 is as follows: The filtered sample image γ (i,j,B) The fixed noise content N of the separated instrument (i,j,B) Subtraction yields the final clear image X′. (i,j,B) ; X′ (i,j,B) =c (i,j,B) -N (i,j,B) Complete the non-uniform noise calibration of hyperspectral data of brain tumor tissue.

Images