Method and system for improving PS-OCT visibility based on polarization multi-parameter fusion

Abstract

The present invention discloses a method and system for improving the visibility of PS-OCT based on polarization multi-parameter fusion. The method includes obtaining an original PS-OCT image and performing preprocessing to obtain the contour of the sample to be measured, filtering a QUV three-dimensional array constructed based on the Stokes matrix using the contour of the sample to be measured to obtain a polarization state image, calculating a local optical axis image and a local phase retardation image using a Poincaré sphere based on the polarization state obtained by calculation, performing average gradient fusion or weighted fusion on the obtained multiple types of images, and obtaining a final PS-OCT image after fusion. This method and system do not require the input of polarized light that interferes with each other, only require a single input polarization state, and have low requirements for the complexity of the system. It has a wide range of applications and may be used in endoscope-based PS-OCT or galvano-meter-based planar scanning PS-OCT.

Description

Technical Field

[0001] The present invention relates to the field of optical coherence tomography technology, and particularly relates to a method and system for improving the visibility of PS-OCT based on polarization multi-parameter fusion.

Background Art

[0002] Optical coherence tomography (OCT) is the highest-resolution imaging method recognized in the current field of endoluminal imaging. However, the conventional OCT system that images based on intensity is insufficient in terms of tissue characteristic analysis. As a result, it is difficult for humans or AI to distinguish different biological tissues such as different membrane structures and plaques when analyzing images. The intensity-based OCT currently used in clinical practice has a blurred image representation result during use. However, in addition to intensity information, light also carries other additional characteristics. By using the additional characteristics carried by light, analysis or quantitative measurement can be performed on different tissues to improve visibility. The principle of polarization-sensitive OCT (PS-OCT) is based on this. Multiple types of biological tissues or samples can achieve polarization modulation of the input light, change the polarization state of the input light, obtain a method for additionally representing the visibility of biological tissues or samples, and obtain characteristic information other than intensity.

[0003] ​​The PS-OCT technology can change the polarization state of the input polarization by propagating and reflecting the polarized light incident on the tissue of the sample in the medium of the sample, and demodulating the polarization state of the back-reflected light, so as to obtain the polarization characteristic information of the sample and realize depth-resolved imaging by the birefringence of the tissue. This special property is particularly important in samples or biological tissues. Proteins or biopolymer substances with isotropic tissue structures in blood vessels, such as collagen and actin, can change the polarization state of the incident light. It is induced to generate form birefringence and generate measurable optical signals. By measuring the polarization state of the back-reflected light or back-scattered light from the sample or biological tissue, polarization information of the sample in depth resolution, such as phase delay and the direction of the optical axis, can be obtained.

[0004] Currently, most of the conventional technologies for calculating PS-OCT polarization information are phase delay algorithms and polarization degree algorithms based on the Jones matrix or Mueller matrix. However, due to obvious attenuation effects and depolarization effects in the system, the polarization visibility calculated by the current algorithms is low. Therefore, a method based on multi-parameter fusion analysis is proposed to calculate polarization characteristic information and improve the visibility of polarization images.

Summary of the Invention

Problems to be Solved by the Invention

[0005] Therefore, the present invention aims to provide a method and system for improving the visibility of PS-OCT based on polarization multi-parameter fusion to realize color images of different tissue structures in a sample, help doctors' diagnosis more intuitively, and reduce the learning difficulty of conventional intravascular imaging products.

Means for Solving the Problems

[0006] To achieve the above object, the method for improving the visibility of PS-OCT based on polarization multi-parameter fusion according to the present invention is S1 acquires the original PS-OCT image, preprocesses the original PS-OCT image, and obtains the contour of the sample awaiting measurement. Using the contours of the samples awaiting measurement, the QUV 3D array constructed based on the Stokes matrix is ​​filtered. Polarization state and S2 obtains a polarization state image. In S2 S3 calculates the local optical axis image and local phase delay image using a Poincaré sphere based on the calculated polarization state. The process includes step S4, in which average gradient fusion or weighted fusion is performed on multiple types of images obtained in S2 and S3, and a final PS-OCT image is obtained after fusion.

[0007] More preferably, in S1, the pretreatment is S101 is a process that applies a cosine taper window to the H-channel and V-channel data of the two polarization states in the original PS-OCT image to shape it. S102 performs a Fourier transform on the reshaped data to obtain the Fourier domain matrices for the H channel and V channel. After taking the average value of the Fourier domain matrices of the H channel and V channel, the original images of the H channel and V channel are fused together in S103. The process includes a step S104 in which noise is filtered and removed from the fused image according to a set threshold to obtain the contour of the sample awaiting measurement.

[0008] More preferably, when the original images of the H channel and V channel are merged in S103, the merging formula is as follows:

number

[0009] More preferably, in S2, the polarization state image is After normalizing the Stokes matrix, a QUV three-dimensional array is formed, and using the contour of the sample waiting to be measured, the QUV three-dimensional array is filtered, pixel points outside the contour are set to 0, the QUV three-dimensional array is drawn in the RGB format, and a polarization state color image of the contour of the sample waiting to be measured is obtained. The method adopted is as described above.

[0010] More preferably, in S3, when calculating the local optical axis image, Using the Poincaré sphere, the spatial normal vector B of the plane of the PS-OCT image n is extracted, the second dimension and the third dimension of B n are exchanged to obtain an x*y*3 matrix, which is filtered through the contour of the sample waiting to be measured to obtain an optical axis image, and further includes that x and y represent the number of rows and columns of the pixels of the image.

[0011] More preferably, the local phase retardation is calculated by adopting the following formula,

Equation

[0012] More preferably, the local optical axis image is calculated by adopting the following formula,

Equation

[0013] More preferably, in S4, average gradient fusion is performed on multiple types of images. The method includes normalizing three types of images—a polarization state image, a local optical axis image, and a local phase delay image—and fusing the computational results of the three images using gradient features and adjustable fusion weight coefficients.

[0014] More preferably, in S4, weighted fusion is performed on multiple types of images. S401 converts three types of images—a polarization state image, a local optical axis image, and a local phase delay image—into a grayscale image. The method includes S402, which involves performing grayscale feature fusion, shape feature fusion, and texture feature fusion on three types of grayscale images, then fusing them again to obtain a final fused image, and finally obtaining a PS-OCT image according to the following fusion formula.

number

[0015] More preferably, in S402, the grayscale feature fusion is For a grayscale image, grayscale feature values ​​including mean, variance, energy, slope, and kurtosis are extracted. A weighted fusion method is employed to fuse the original three images into a figure based on five aforementioned grayscale feature values. The fused image of the five grayscale feature values ​​is calculated for each, This process includes fusing the five fusing image files of grayscale features together to create a single final grayscale-fused image.

[0016] More preferably, in S402, the shape feature fusion includes extracting shape features from three types of grayscale images, normalizing the central moment to obtain seven invariant moment shape features, using the seven shape features as a shape feature vector to form a shape feature matrix, performing shape fusion using the shape feature matrix to obtain a shape fused image.

[0017] More preferably, in S402, the texture feature fusion includes extracting texture features, including energy, entropy, contrast, and association, from three types of grayscale images, constructing a texture feature vector using the texture features, fusing the three types of grayscale images according to the constructed four texture feature vectors to form four texture feature images, and then fusing the four texture feature images according to equal weights to form a texture fusion image.

[0018] The present invention is used to implement the above-described method for improving PS-OCT visibility based on polarization multi-parameter fusion. An image acquisition module for obtaining the original PS-OCT image, performing preprocessing on the original PS-OCT image, and obtaining the contour of the sample awaiting measurement. An image processing module for using the contours of samples awaiting measurement to filter a QUV 3D array constructed based on a Stokes matrix to obtain a polarization state image, and then using a Poincaré sphere to calculate a local optical axis image and a local phase delay image based on the calculated polarization state. The present invention further provides a PS-OCT visibility improvement system based on polarization multi-parameter fusion, comprising an image fusion module for performing average gradient fusion or weighted fusion on the multiple types of images obtained above, and obtaining a final PS-OCT image after fusion. [Effects of the Invention]

[0019] The PS-OCT visibility improvement method and system based on polarization multi-parameter fusion disclosed in this application have at least the following advantages compared to the prior art.

[0020] 1. The PS-OCT visibility improvement method and system based on polarization multi-parameter fusion according to the present application does not require mutually interfering polarization inputs during PS-OCT acquisition, but requires only one single input polarization state, thus having a low requirement for system complexity. It has a wide range of applications and may be used in endoscope-based PS-OCT or galvanometer-based planar scanning PS-OCT.

[0021] 2. The PS-OCT visibility improvement method and system based on polarization multi-parameter fusion according to the present application employs multiple methods to calculate and fuse polarization information, resulting in higher visibility compared to existing technologies. [Brief explanation of the drawing]

[0022] [Figure 1] This is a flowchart of the PS-OCT visibility improvement method based on polarization multi-parameter fusion according to the present invention. [Figure 2] This is a schematic diagram of the PS-OCT image interface in the PS-OCT visibility improvement method based on polarization multi-parameter fusion according to the present invention. [Figure 3] This is a schematic diagram of a Poincaré sphere in the PS-OCT visibility improvement method based on polarization multi-parameter fusion according to the present invention. [Figure 4] This is a schematic diagram of the local optical axis in the PS-OCT visibility improvement method based on polarization multi-parameter fusion according to the present invention. [Figure 5] This is a schematic diagram of the structure of the PS-OCT visibility improvement system based on polarization multi-parameter fusion according to the present invention. [Modes for carrying out the invention]

[0023] The present invention will be described in further detail below with reference to the drawings and specific embodiments.

[0024] As shown in Figure 1, the PS-OCT visibility improvement method based on polarization multi-parameter fusion according to one embodiment of the present invention is S1 acquires the original PS-OCT image, preprocesses the original PS-OCT image, and obtains the contour of the sample awaiting measurement. Using the contours of the samples awaiting measurement, the QUV 3D array constructed based on the Stokes matrix is ​​filtered. Polarization state and S2 obtains a polarization state image. profit S3 calculates a local optical axis image and a local phase delay image using a Poincaré sphere based on the determined polarization state. The process includes step S4, in which average gradient fusion or weighted fusion is performed on multiple types of images obtained in S2 and S3, and a final PS-OCT image is obtained after fusion.

[0025] In one embodiment of the present application, the pretreatment process is as follows: S101 is a process that applies a cosine taper window to the H-channel and V-channel data of the two polarization states in the original PS-OCT image to shape it. S102 performs a Fourier transform on the reshaped data to obtain the Fourier domain matrices for the H channel and V channel. After taking the average value of the Fourier domain matrices of the H channel and V channel, the original images of the H channel and V channel are fused together in S103. The process includes a step S104 in which noise is filtered and removed from the fused image according to a set threshold to obtain the contour of the sample awaiting measurement.

[0026] In S103, when fusing the original images of the H channel and V channel, the fusing formula is as follows:

number

[0027] In the process of implementing this, the system is in a state of two orthogonal polarization states. H channel and V channel A-scan Day Ta The images are collected, a reference plane is drawn over them, a cosine taper window is applied to shape them, variance compensation is added, and then an FFT is performed to obtain the Fourier domains IMG_H and IMG_V for the H and V channels. Since these are complex matrices of x*y*4, when displaying them as images, their absolute values ​​are taken and the average is taken along the third dimension to obtain the original images for the H and V channels. Due to the birefringence effect of polarization, a phase delay occurs, so as shown in Figure 2, there are two images, one above and one below, for a total of four images. By fusing these four images, a structural diagram of the sample awaiting measurement can be obtained, and the fusing formula is as follows:

number

[0028] In S2, the polarization state image is The method employed involves normalizing the Stokes matrix to form a QUV 3D array, filtering the QUV 3D array using the contour of the sample awaiting measurement, setting pixels outside the contour to 0, and then rendering the QUV 3D array in RGB to obtain a color image of the polarization state of the contour of the sample awaiting measurement.

[0029] The polarization state is S0=e 2x +e2y S1=e 2x -e 2y S² = 2e x e y cosθ, S3 = 2e x e y This is expressed using the Stokes parameter sinθ. Of course, S0 2 =S1 2 +S2 2 +S3 2 Therefore, in the equation, only three variables are independent, and in the ideal case (i.e., lossless transmission), S0 = constant, so S1, S2, and S3 represent a single sphere. The sphere with S0 = 1 is called the Poincaré sphere, and each point on the sphere corresponds one-to-one with the total polarization state of light. The following is the calculation process for the polarization states.

[0030] In Step 1,

number

[0031] In step 2, normalize, Q = S1 / S0, U = S2 / S0, and V = S3 / S0, where Q, U, and V are normalized coordinates.

[0032] In step 3, filtering is performed. The filtering method may be, but is not limited to, median filtering, Gaussian filtering, mean filtering, imbox filtering, Wiener filtering, dilation and deflation, etc.

[0033] In step 4, construct a 3D array. Q, U, and V are used to construct a 3D array of x*y*3 (Stokes array). VectorAfter construction, Stokes is filtered using the contour Msk_Thr array, pixels outside the contour are set to 0, and finally the resulting 3D array is plotted using the RGB method to obtain a color image of the polarization state of the contour of the sample awaiting measurement.

[0034] As shown in Figures 2-3, when calculating the local optical axis image in S3, Using a Poincaré sphere, the spatial binormal vector B of the PS-OCT image plane. n Extract B n The second and third dimensions are swapped to obtain an x*y*3 matrix, which is filtered through the contour of the sample awaiting measurement to obtain an optical axis image, in which x and y represent the number of rows and columns of pixels in the image.

[0035] Based on the above polarization state, Stokes is used. Vector This may be represented by a Poincaré sphere. In Figure 3, P1, P2, and P3 are three polarization states represented by the Stokes parameters (S1, S2, S3), which lie on a sphere, and plane a is a plane fitted by the three points P1, P2, and P3, and A1 is the plane normal vector of a, i.e., the optical axis required here. P1 is the incident polarization state or input polarization state, and P1 is incident on the surface of the sample, reflected directly from the surface to become the output polarization state, and since the polarization information does not change, P1 is also an output polarization state. P2 and P3 are obtained by rotating P1 by a certain angle around the optical axis A1 of the sample. P1, P2, and P3 are the output polarization states received by the balance detector.

number

[0036] B n The method for calculating the optical axis based on this is as follows:

[0037] First, B n By swapping the second and third dimensions, we obtain an x*y*3 matrix (where x and y represent the number of rows and columns of pixels in the image), and then filtering it with Msk_Thr, we can obtain an optical axis image.

[0038] Furthermore, the local phase delay is calculated using the following formula:

number

[0040] B n The optical axis represented by this is the result of superimposing the birefringence effects of tissues at different depths. Deeper in the tissue, the accumulated birefringence effect causes distortion of the result. Therefore, unless the accumulated birefringence effect is removed according to depth, the true optical axis information deep within the tissue will be lost. The following formula A in n It is not possible to restore a local optical axis like this.

[0041] Local optical axis A n The calculation process is as follows:

number

[0042] In S4, performing average gradient fusion on multiple types of images is possible. Three types of images—polarization state image, local optical axis image, and local phase delay image—are normalized. This method involves fusing the results of calculations on three images using gradient features and adjustable fusion weight coefficients.

[0043] First, the results of the three calculation methods are normalized and the calculations are performed within the same range.

number

[0044] The formula for calculating the gradient feature vector G is as follows:

number

number

number

[0045] In S4, performing weighted fusion on multiple types of images is possible. S401 converts three types of images—a polarization state image, a local optical axis image, and a local phase delay image—into a grayscale image. The method includes S402, which involves performing grayscale feature fusion, shape feature fusion, and texture feature fusion on three types of grayscale images, then fusing them again to obtain a final fused image, and finally obtaining a PS-OCT image according to the following fusion formula.

number

[0046] In S402, the grayscale feature fusion is performed as follows: For a grayscale image, grayscale feature values ​​including mean, variance, energy, slope, and kurtosis are extracted. A weighted fusion method is employed to fuse the original three images into a figure based on five aforementioned grayscale feature values. The fused image of the five grayscale feature values ​​is calculated for each, This process includes fusing the five fusing image files of grayscale features together to create a single final grayscale-fused image.

[0047] In S402, the shape feature fusion includes extracting shape features from three types of grayscale images, normalizing the central moment to obtain seven invariant moment shape features, using the seven shape features as a shape feature vector to form a shape feature matrix, performing shape fusion using the shape feature matrix, and obtaining a shape fused image.

[0048] In S402, the texture feature fusion includes extracting texture features, including energy, entropy, contrast, and association, from three types of grayscale images, constructing a texture feature vector using the texture features, fusing the three types of grayscale images according to the four constructed texture feature vectors to form four texture feature images, and then fusing the four texture feature images according to equal weights to form a texture fusion image.

[0049] 1) Tonal characteristics a. Extraction of grayscale features The grayscale characteristics are: mean m, variance v 2 It includes five types of statistics: energy e, slope s, and kurtosis u. The meanings and calculation formulas for these five types of statistics are as follows.

[0050] first,

number

[0051] The average value m represents the average energy of the image, and the calculation formula is as follows:

number

number

number

number

number

[0052] b. Tonal feature fusion Based on five types of grayscale features, a weighted fusion method is adopted, and the results of three calculation methods are fused into five feature value diagrams, with the feature vector denoted by h, where h = (m,v 2 The three feature vectors are denoted by h1, h2, and h3 respectively, and the weights of each feature are as follows:

number

[0053] F: A fused image of five features j The following formulas are obtained by calculating each of the following:

number

[0054] The five fused feature quantities are fused again to obtain a single grayscale image, and the formula is as follows:

number

[0055] 2) Shape characteristics a. Extraction of shape features For a discrete digital image P(x,y), the p+q-th standard moment of the image is given by:

number

[0056] The central moment of order p+q is shown as follows:

number

number

number

number

number

[0057] b. Fusion of shape features The shape feature vector is denoted by M, where M = (φ1, φ2, φ3, φ4, φ5, φ6, φ7).

[0058] A matrix constructed by representing the shape feature vectors of the three types of calculation results as M1, M2, and M3 is shown as follows:

number

number

[0059] 3) Texture Features a. Extraction of texture features Texture features are extracted based on a simultaneous tone generation matrix. A tone histogram can directly describe the tone distribution of a single pixel, while a simultaneous tone generation matrix can describe the combined tone distribution of two pixels. Let (x,y) be a point in an image, and let (gx,gy) be its tone distribution. When (x,y) moves, a point (x+i,y+j) is obtained, and the corresponding (gx',gy') is also generated. The number of times each tone value appears in a single image is statistically calculated, all tone values ​​are constructed into a single square matrix, and the number of occurrences of a given tone value and the total number of occurrences are normalized to obtain P(gx,gy), i.e., the probability of occurrence, which is called the simultaneous tone generation matrix. The normalization formula for the simultaneous tone generation matrix is ​​as follows, where Z represents the width and height of the square matrix image, i.e., the image size is Z×Z.

number

[0060] Since the simultaneous occurrence matrix alone cannot fully represent the texture features of an image, four scalars—energy, entropy, contrast, and association—are introduced to supplementarily represent the texture features of the image. The meanings and calculation methods of these four scalars are as follows.

[0061] Regarding energy E, the energy value not only describes the uniformity of the grayscale distribution but can also express to some extent the roughness of the texture. When the numerical values ​​of all parameters in the grayscale simultaneous generation matrix P(gx,gy) are equal, E is relatively small, and when the magnitudes of the parameter values ​​are clearly distinguishable, E is large. When the parameters in the grayscale simultaneous generation matrix are close to the center, E is large, which indicates that the texture of the image is well uniform and the changes are regular. The calculation method is to first square each parameter of the elements in the grayscale simultaneous generation matrix P(gx,gy) and then add them together, as shown in the formula below.

number

[0062] Entropy S is used to indicate the complexity of an image's texture, and it is large when the values ​​of the co-occurrence matrix are relatively uniform. The formula for calculating it is as follows:

number

[0063] Regarding contrast I, contrast is used to express the degree of image sharpness. The lower the contrast, the shallower the grooves that represent the degree of surface irregularities of the object, and the lower the degree of image sharpness. The calculation formula is as follows:

number

[0064] Regarding the correlation R, it is used to indicate the equality of the parameters of the grayscale simultaneous occurrence matrix P(gx,gy) in the horizontal and vertical directions. If the correlation value in one direction is greater than that of another, the texture features in that direction are clearer, and therefore the correlation allows us to find directions where the texture is relatively stronger. The calculation formula is as follows:

number

number

[0065] b. Texture feature fusion A texture feature vector Y=(E,S,I,R) is constructed for energy E, entropy S, contrast I, and relevance R. The feature vectors of the three resulting images are Y1, Y2, and Y3, respectively, and the formula for calculating the weights is as follows.

number

[0066] First, the different texture features of the three images were merged to obtain four types of fused feature images, and the fusion formula is as follows:

number

[0067] Finally, the four feature images were fused, and the fusion formula is as follows:

number

[0068] 4) Feature-weighted fusion The above three characteristics yield the fusion results of gradation, shape, and texture, respectively. These three characteristics are then fused again to obtain the final fused image. The fusion formula is as follows:

number

number

[0069] As shown in Figure 5, the present invention further provides a PS-OCT visibility improvement system based on polarization multiparameter fusion for implementing the above-described PS-OCT visibility improvement method based on polarization multiparameter fusion, which is: An image acquisition module for obtaining the original PS-OCT image, performing preprocessing on the original PS-OCT image, and obtaining the contour of the sample awaiting measurement. Using the contours of the samples awaiting measurement, the QUV 3D array constructed based on the Stokes matrix is ​​filtered. Polarization state, Polarization state images, polarization degree DOPU images, and phase delay DPPR images were obtained. , profit An image processing module for calculating an optical axis image, a local optical axis image, and a local phase delay image using a Poincaré sphere based on the determined polarization state, The system includes an image fusion module for performing average gradient fusion or weighted fusion on the multiple types of images obtained above, and obtaining a final PS-OCT image after fusion.

[0070] Of course, the above embodiments are merely examples given for illustrative purposes and do not limit the embodiments. Those skilled in the art can make further variations or modifications based on the above description. There is no need or method to list all embodiments here. Any obvious variations or modifications resulting therefrom are also protected within the scope of the invention.

Claims

1. S1 acquires the original PS-OCT image, preprocesses the original PS-OCT image, and obtains the contour of the sample awaiting measurement. S2 uses the contour of the sample awaiting measurement to filter the QUV 3D array constructed based on the Stokes matrix and obtain the polarization state and polarization state image. S3 calculates a local optical axis image and a local phase delay image using a Poincaré sphere based on the obtained polarization state. The process includes step S4, in which average gradient fusion or weighted fusion is performed on multiple types of images obtained in S2 and S3, and a final PS-OCT image is obtained after fusion. In S3, when calculating the local optical axis image, This method further includes using a Poincaré sphere to extract the spatial binormal vector Bn of the PS-OCT image plane, swapping the second and third dimensions of Bn to obtain an x*y*3 matrix, filtering it through the contour of the sample awaiting measurement to obtain an optical axis image, where x and y represent the number of rows and columns of pixels in the image. A method for improving PS-OCT visibility based on polarization multi-parameter fusion, characterized by the features described above.

2. In S1, the pretreatment is performed as follows: S101 is a process in which a cosine taper window is applied to the H channel data and V channel data of the two polarization states in the original PS-OCT image to reshape them. S102, a Fourier transform is performed on the formatted data to obtain the Fourier domain matrices for the H channel and V channel. After taking the average value of the Fourier domain matrices of the H channel and V channel, the original images of the H channel and V channel are fused together as the original images of the H channel and V channel in S103. The process includes step S104, which involves filtering and removing noise from the fused image according to a set threshold to obtain the contour of the sample awaiting measurement. A method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 1.

3. In S103, when fusing the original images of the H channel and V channel, the fusing formula is as follows: [Math 1] Str ｔｏｔａｌ This is the image after fusion, and positive １ pH ２ These are the two images, upper and lower, of the H channel, respectively, and pV １ and pV ２ These are the two images, one above and one below the V channel. The method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 2.

4. In S2, the polarization state image is The method employed involves normalizing the Stokes matrix, forming a QUV 3D array, filtering the QUV 3D array using the contour of the sample awaiting measurement, setting pixels outside the contour to zero, and then plotting the QUV 3D array in RGB to obtain a color image of the polarization state of the contour of the sample awaiting measurement. A method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 1.

5. The local phase delay in the aforementioned local phase delay image is calculated using the following formula: [Math 2] δ ｎ This is the local phase delay, and N ｎ This is the normal vector in the nth contact plane, and N ｎ－１ This is the normal vector of the (n-1)th contact plane. A method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 1.

6. The aforementioned local optical axis image is calculated using the following formula: [Math 3] A ｎ is the local optical axis, and B ｎ represents the optical axis obtained by superimposing the birefringence effects of tissues at different depths. R ｎ is the 3×3 rotation matrix from the (n−1)-th optical axis to the n-th optical axis. δ ｎ is the phase retardation of the n-th contact plane. A ｎ (x), A ｎ (y), A ｎ (z) are the three dimensions of the local optical axis A of the three-dimensional array, respectively, ｎ ​ The method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 5.

7. Performing average gradient fusion on multiple types of images is, This method includes normalizing three types of images—a polarization state image, a local optical axis image, and a local phase delay image—and fusing the computational results of the three images using gradient features and adjustable fusion weight coefficients. A method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 1.

8. Performing weighted fusion on multiple types of images is, S401 converts three types of images—a polarization state image, a local optical axis image, and a local phase delay image—into a grayscale image. The method includes S402, which involves performing grayscale feature fusion, shape feature fusion, and texture feature fusion on three types of grayscale images, then fusing them again to obtain a final fused image, and then obtaining a final PS-OCT image according to the following fusion formula. [Math 4] d ｉ This is the fusion coefficient, Fusimage ｉ This shows the image after grayscale feature fusion, shape feature fusion, and texture feature fusion. A method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 1.

9. The aforementioned grayscale feature fusion is For a grayscale image, grayscale feature values ​​including mean, variance, energy, slope, and kurtosis are extracted. A weighted fusion method is employed, and the original three images are fused into a figure based on five aforementioned grayscale feature values. The fused image of the five grayscale feature values ​​is calculated for each, This process includes merging the five merged tonal feature value images again to ultimately create a single tonal fused image. The method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 8.

10. The aforementioned shape feature fusion involves extracting shape features from three types of grayscale images, normalizing the central moment, obtaining seven invariant moment shape features, using the seven shape features as a shape feature vector, forming a shape feature matrix, performing shape fusion using the shape feature matrix, and obtaining a shape-fused image. The method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 8.

11. The aforementioned texture feature fusion includes extracting texture features, including energy, entropy, contrast, and association, from three types of grayscale images, constructing a texture feature vector using the texture features, fusing the three types of grayscale images according to the four constructed texture feature vectors to form four texture feature images, and then fusing the four texture feature images according to equal weights to form a texture fusion image. The method for improving PS-OCT visibility based on polarization multi-parameter fusion as described in feature 8.

12. Used to implement the PS-OCT visibility improvement method based on polarization multi-parameter fusion described in any one of claims 1 to 11 above, An image acquisition module for acquiring the original PS-OCT image, performing preprocessing on the original PS-OCT image, and obtaining the contour of the sample awaiting measurement. An image processing module for filtering a QUV 3D array constructed based on a Stokes matrix using the contour of a sample awaiting measurement, obtaining polarization states and polarization state images, and calculating local optical axis images and local phase delay images using a Poincaré sphere based on the obtained polarization states, The system includes an image fusion module for performing average gradient fusion or weighted fusion on the multiple types of images obtained above, and obtaining a final PS-OCT image after fusion. When the image processing module calculates the local optical axis image, This method further includes using a Poincaré sphere to extract the spatial binormal vector Bn of the PS-OCT image plane, swapping the second and third dimensions of Bn to obtain an x*y*3 matrix, filtering it through the contour of the sample awaiting measurement to obtain an optical axis image, where x and y represent the number of rows and columns of pixels in the image. A PS-OCT visibility improvement system based on polarization multi-parameter fusion, characterized by the above.

Images