Glaze pattern structural feature extraction method based on polarization difference and band fusion

Abstract

The application discloses a glaze pattern structure feature extraction method based on polarization difference and waveband fusion, and belongs to the technical field of image processing. A plurality of images of the same pattern under different polarization angles and different illumination wavebands are collected, and first, the plurality of images are registered; then, polarization parameters are calculated and highlight areas are detected; then, based on a polarization degree physical model, specular reflection components are estimated and suppressed to obtain a highlight-removed pattern image; subsequently, the image is subjected to color and scale normalization processing; finally, a pattern skeleton is segmented and thinned from the normalized image, color, curve, period and texture features are extracted, and structured feature codes are fused and generated. The application can effectively suppress highlight interference, improve the definition and readability of pattern details, realize the consistency and reproducibility of data under different collection conditions, and output searchable structured features, thereby greatly improving the efficiency of digital recording, management and research of glaze patterns.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a method for extracting structured features of glaze patterns based on polarization difference and band fusion. Background Technology

[0002] Current methods for digitally acquiring and analyzing glazed ceramic patterns typically rely on ordinary cameras / mobile phones for shooting, combined with external lighting, polarizing filters, or light booths. After acquisition, the images are then manually edited or processed using general highlight removal algorithms. However, due to the strong specular reflection of the glaze, the highlight position shifts rapidly with changes in lighting direction and shooting angle, often obscuring fine lines, gilding, and dark patterns. Furthermore, differences in the color temperature of light sources at different locations, as well as variations in exposure strategies and shooting distances, cause color and scale to become non-standardized, resulting in inconsistent results for the same pattern under different acquisition conditions, making it difficult to reproduce and compare across different locations.

[0003] Existing de-reflection / de-highlight methods mainly rely on single-frame image threshold restoration or simple fusion, lacking a mechanism for reliable estimation and suppression of specular reflection components based on imaging condition differences such as polarization angle differences and multi-band differences. This easily leads to the accidental erasure of pattern details, boundary distortion, or residual highlights. Furthermore, existing processes often only output enhanced images, lacking structured analysis and searchable encoding of pattern elements such as edge period, curve skeleton, and hierarchical structure, making it difficult to meet the technical requirements of "reproducible, comparable, and searchable" for cultural relic recording, pattern research, and digital management.

[0004] Therefore, how to provide a method for extracting structured features of glaze patterns that can improve the consistency, reliability, and application efficiency of pattern data is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a method for extracting structured features of glaze patterns based on polarization difference and band fusion to overcome or at least partially solve the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] This invention provides a method for extracting structured features of glaze patterns based on polarization difference and band fusion, comprising the following steps:

[0008] S1: Obtain a set of original images of the same glaze pattern area under multiple imaging conditions; the multiple imaging conditions include at least different polarization angle sequences and different illumination band sequences;

[0009] S2: Perform image registration on the original image set so that the same physical point in all images is mapped to the same pixel coordinate, and obtain the registered image set;

[0010] S3: Based on the images with different polarization angles in the registered image set, calculate the polarization parameters of each pixel;

[0011] S4: Generate a binary mask that identifies the specular highlight area based on the polarization parameters and pixel intensity;

[0012] S5: Based on the polarization parameters and the binary mask, estimate and remove the specular reflection component from the registered image set to obtain a de-spectral textured image;

[0013] S6: Perform color channel correction on the de-highlighted texture image and output a standard texture image after color and scale normalization;

[0014] S7: Perform pattern region segmentation and skeletonization processing on the standard pattern image to extract the geometric skeleton of the pattern;

[0015] S8: Based on the standard pattern image and the geometric skeleton, extract and fuse color, curve, period and texture features to generate the structured features of the glaze pattern.

[0016] Preferably, in step S1, the different polarization angle sequence includes images with polarization directions of 0°, 45°, 90° and 135°; the different illumination band sequence includes images acquired under illumination from at least two different wavelength light sources.

[0017] Preferably, step S2 specifically includes:

[0018] S21. Select one frame from the original image set as a reference frame;

[0019] S22. Extract feature points between the remaining frames and the reference frame and match them to obtain a set of matching point pairs;

[0020] S23. Based on the set of matching point pairs, solve for the parameters of the affine transformation model using a robust estimation algorithm;

[0021] S24. Using the affine transformation model parameters, transform the remaining frames of images to the coordinate system of the reference frame to complete the registration.

[0022] Preferably, in step S3, the polarization parameters include Stokes parameters. , , The formula for calculating the Stokes parameter is as follows:

[0023] ;

[0024] ;

[0025] ;

[0026] in, , , , These represent the intensity values ​​at pixel x of the registered images with polarization angles of 0°, 45°, 90°, and 135°, respectively.

[0027] Preferably, the linear polarization degree DoLP and polarization angle AoP of each pixel are calculated based on the Stokes parameters, using the following formula:

[0028] ;

[0029] ;

[0030] in It is the tangent of the four quadrants. It refers to the degree of polarization of the pixel. It is the main polarization direction of the pixel. It is a very small positive number.

[0031] Preferably, step S4 specifically includes:

[0032] Calculate the local brightness adaptive threshold for pixel x:

[0033] ;

[0034] in, , They are respectively based on The mean and standard deviation of the brightness within a local window centered on the α value are given by α, which is a preset empirical coefficient.

[0035] Define specular candidate mask for:

[0036] ;

[0037] ;

[0038] Where L(x) is the intensity of pixel x, This is the preset polarization threshold.

[0039] Preferably, step S5 specifically includes:

[0040] S51. Select the image with the lowest intensity of each pixel from the registered image set. As a base image;

[0041] S52. Estimate the specular reflection component based on the linear polarization degree DoLP(x) and intensity L(x):

[0042] ;

[0043] Where β is a preset weighting coefficient;

[0044] S53. Within the area marked as a highlight by the binary mask, from the base image... Subtract the specular reflection component from the middle The results are then truncated to obtain a de-highlight texture image. :

[0045] ;

[0046] in, It is a truncation function.

[0047] Preferably, step S6 specifically includes:

[0048] S61: Regarding the dehighlight texture image For each color channel c∈{R, G, B}, calculate its average intensity outside the highlight region. ;

[0049] S62: Calculate the gain coefficient for each channel:

[0050] ;

[0051] in, The target grayscale mean is... It is a very small positive number;

[0052] S63: Convert the pixel values ​​of each channel Multiply by the corresponding gain coefficient to obtain the color-normalized image. :

[0053] .

[0054] Preferably, step S7 specifically includes:

[0055] S71. Calculate the gradient magnitude of the standard pattern image, and obtain a binary mask of the pattern region through threshold segmentation. ;

[0056] S72, the binary mask A thinning algorithm is applied to iteratively delete boundary pixels that meet the preset deletion conditions until a single-pixel-wide pattern skeleton image Sk is obtained.

[0057] Preferably, in step S8, the structured feature F is composed of the following sub-feature vectors concatenated:

[0058] Color characteristics , which is the color histogram of the standard pattern image;

[0059] Curve characteristics It is a histogram of the orientation angles of each pixel on the geometric skeleton;

[0060] Periodic characteristics It is obtained by calculating the autocorrelation function of the one-dimensional projection signal along the pattern border direction and extracting the period length L corresponding to the main peak position;

[0061] Texture features It is a local binary pattern histogram of the standard pattern image.

[0062] The technical solution provided in this invention addresses high-reflectivity glaze pattern images. It utilizes data acquired under multiple polarization angles to suppress specular reflection, employs image acquisition data under multi-wavelength illumination to achieve stable separation of pattern images under complex lighting conditions, and utilizes feature point matching from multiple frames to achieve adaptive image registration. Finally, it completes image color and scale normalization and outputs structured pattern feature codes. The beneficial effects of this technical solution include at least:

[0063] This invention, through multi-polarization angle image acquisition and specular reflection component estimation and suppression algorithm, can significantly reduce the obstruction of glaze highlights, improve the clarity and readability of pattern details, and reduce the workload of repeated debugging and post-processing.

[0064] This invention standardizes color and scale, improves data consistency and reproducibility under different site and operator conditions, and facilitates long-term archiving, horizontal comparison and research analysis.

[0065] This invention can output standardized pattern diagrams and structured feature codes, supporting pattern classification, similarity retrieval and digital management, and improving the efficiency of pattern research and cultural relic recording. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0067] Figure 1 A flowchart of the method for extracting structured features of glaze patterns based on polarization difference and band fusion provided in an embodiment of the present invention;

[0068] Figure 2 This is a flowchart of the process for obtaining a de-highlight textured image provided in an embodiment of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] This invention discloses a method for extracting structured features of glaze patterns based on polarization difference and band fusion, such as... Figure 1 As shown. It includes the following steps:

[0071] S1: Obtain the original image set of the same glaze pattern area under multiple image composition conditions; the multiple image composition conditions include at least different polarization angle sequences and different illumination band sequences;

[0072] S2: Perform image registration on the original image set so that the same physical point in all images is mapped to the same pixel coordinate, and obtain the registered image set;

[0073] S3: Calculate the polarization parameters of each pixel based on images with different polarization angles in the registered image set;

[0074] S4: Generate a binary mask that identifies the specular highlight area based on polarization parameters and pixel intensity;

[0075] S5: Based on polarization parameters and a binary mask, specular reflection components are estimated and removed from the registered image set to obtain a de-spectral textured image;

[0076] S6: Perform color channel correction on the de-highlighted texture image and output a standard texture image after color and scale normalization;

[0077] S7: Perform pattern region segmentation and skeletonization processing on the standard pattern image to extract the geometric skeleton of the pattern;

[0078] S8: Based on standard pattern images and geometric skeletons, extract and fuse color, curve, period and texture features to generate structured features of glaze patterns.

[0079] In this embodiment, for scenarios involving the recording and design research of cultural relics that require obtaining pattern information from highly reflective surfaces such as glazed ceramics, high-quality, high-definition pattern acquisition and stable de-reflection output of samples with different glaze colors and reflectivity intensities can be achieved by collecting different set values. These different set values ​​include polarization angle sequence, illumination band and brightness, and working distance threshold.

[0080] In one embodiment, in step S1, the different polarization angle sequences include images with polarization directions of 0°, 45°, 90°, and 135°; the different illumination band sequences include images acquired under illumination from at least two different wavelength light sources.

[0081] In one embodiment, step S2 transforms images with different polarization angles or different wavelengths to the same coordinate system through multi-frame registration, ensuring that the same physical point corresponds to the same pixel position in each frame. This resolves slight displacement caused by hand shake and avoids ghosting during subsequent image differencing and fusion. Specifically, it includes:

[0082] S21. Select one frame from the original image set as the reference frame;

[0083] S22. Extract the feature points between each of the remaining frames and the reference frame and match them to obtain a set of matching point pairs;

[0084] S23. Solve the affine transformation model parameters based on the set of matching point pairs using a robust estimation algorithm;

[0085] S24. Using the affine transformation model parameters, transform the remaining frames of images to the coordinate system of the reference frame to complete the registration.

[0086] The execution process of this embodiment is as follows:

[0087] First, determine the reference frame. With the frame to be registered Here, k represents different polarization angles or wavebands. For cases where the pattern is located in an approximate plane and the viewing angle changes little across multiple frames, an affine transformation model is chosen to achieve the geometric transformation of the frames to be registered, ultimately yielding the registered image. Assuming the reference frame pixel coordinates are x=(u, v), then we have the affine equation:

[0088] ;

[0089] in, , These are the mapped coordinates, a matrix. It is a linear parameter matrix that can approximate the small rotations of the camera around the optical axis and the minute changes in distance. It is the translation parameter matrix.

[0090] In reference frame and Feature points are extracted and matched to obtain paired point sets.

[0091] ;

[0092] Expand the affine equation into linear form:

[0093] ;

[0094] Each pair of points yields two equations. Since there are six unknowns, at least three pairs of points must be selected. This can be written in matrix least squares form as follows:

[0095] ;

[0096] It is abbreviated as = Then the least squares solution is:

[0097]

[0098] Because glaze highlights, texture repetition, and noise can introduce erroneous matching points, robust estimation is incorporated. Three sample sets are randomly selected from the matching points, the candidate affine transformation equation T is calculated, and then the reprojection error for each matching point is calculated based on T.

[0099] ;

[0100] like Less than the interior point threshold Then, we select the interior points. Through multiple iterations, we select the affine transformation equation with the most interior points and perform least squares processing.

[0101] In one embodiment, in step S3, when the input includes four polarization angles (0°, 45°, 90°, 135°), the linear polarization Stokes parameters can be calculated. Four frames are used to construct a polarization state description of the pixels, providing a quantitative basis for subsequent specular highlight / diffuse reflection separation. The polarization parameters include Stokes parameters. , , , It is the total intensity term related to brightness. , It is a polarization difference term, reflecting the difference between polarization direction and intensity.

[0102] The formula for calculating the Stokes parameter is:

[0103] ;

[0104] ;

[0105] ;

[0106] in, , , , These represent the intensity values ​​at pixel x of the registered images with polarization angles of 0°, 45°, 90°, and 135°, respectively.

[0107] In one embodiment, the linear polarization degree DoLP and polarization angle AoP of each pixel are calculated based on Stokes parameters. A larger DoLP indicates that the pixel is more likely to contain specular reflections. AoP provides the polarization direction and can be used to further constrain specular component estimation or quality detection. The calculation formula is:

[0108] ;

[0109] ;

[0110] in, It is the tangent of the four quadrants. It refers to the degree of polarization of the pixel. It is the main polarization direction of the pixel. For a very small positive number, 10 can be taken. -6 As a typical value, it is used to improve numerical stability and avoid initial values ​​of 0 in the denominator.

[0111] In one embodiment, in step S4, through the specular candidate mask The highlight area is obtained. In texture recognition, relying solely on polarization can sometimes misidentify certain glazes / textures; therefore, a combined approach of polarization and intensity is commonly used. Specifically, this includes:

[0112] Calculate the local brightness adaptive threshold for pixel x:

[0113] ;

[0114] in, , They are respectively based on The mean and standard deviation of the brightness within a local window centered on the α value are given by α, which is a preset empirical coefficient.

[0115] Define specular candidate mask for:

[0116] ;

[0117] ;

[0118] Where L(x) is the intensity of pixel x, The preset polarization threshold and the manually set polarization threshold are used to determine whether the polarization degree is high enough to distinguish between the diffuse reflection area and the specular highlight area of ​​a light-colored glaze. Although the diffuse reflection of the glaze is relatively bright, its polarization characteristics are weak. When acquiring multiple images with different polarization angles, if the incident light is originally unpolarized, rotating the polarizer will result in almost no change in the transmitted intensity; however, if the light is polarized, the transmitted intensity will change significantly with the angle. Therefore, this method can identify the most likely specular highlight area, and subsequent specular component estimation and suppression can be performed only in... The area is treated to avoid over-processing that would smooth out the pattern.

[0119] In one embodiment, specular reflection often leads to an increase in polarization degree; therefore, step S5 uses DoLP as a proxy for the mirror ratio. Specifically, this includes:

[0120] S51. Select the image with the lowest intensity of each pixel from the registered image set. As a base image;

[0121] S52. Estimate the specular reflection component based on the linear polarization degree DoLP(x) and intensity L(x):

[0122] ;

[0123] Where β is a preset weighting coefficient; Approximating the specular reflection intensity, the higher the intensity and the greater the degree of polarization, the larger the specular component. Simultaneously, through... Limit the processing area to prevent the entire image from turning gray;

[0124] S53. By using a low-highlight image as the base and then subtracting the estimated specular component, more stable texture details can be obtained. Therefore, in the areas marked as highlights by the binary mask, a low-highlight image is selected as the base image: from the base image... Subtract the specular reflection component The results are then truncated to obtain a de-highlight texture image. :

[0125] ;

[0126] ;

[0127] in, It is a truncation function that prevents negative values ​​or values ​​from exceeding the range.

[0128] In one embodiment, step S6 specifically includes:

[0129] S61: Performs gain correction on each of the RGB three channels to eliminate overall color cast caused by different light source color temperatures / exposures, resulting in more stable color characteristics. Specifically for images with de-highlight textures. For each color channel c∈{R, G, B}, calculate its average intensity outside the highlight region. ;

[0130] S62: Calculate the gain coefficient for each channel:

[0131] ;

[0132] in, The target grayscale mean is... For extremely small positive numbers, 10 can also be taken. -6 As a typical value;

[0133] S63: Convert the pixel values ​​of each channel Multiply by the corresponding gain coefficient to obtain the color-normalized image. :

[0134] .

[0135] By implementing built-in calibration and normalization, a stable correspondence between the collected parameters and the output results is achieved, thereby improving the reproducibility and comparability of pattern data.

[0136] In one embodiment, step S7 specifically includes:

[0137] S71. Calculate the gradient magnitude of the standard pattern image and obtain a binary mask of the pattern region through threshold segmentation. ;

[0138] S72, Binary Mask A thinning algorithm is applied to iteratively delete boundary pixels that meet the preset deletion conditions until a single-pixel-wide pattern skeleton image Sk is obtained.

[0139] The specific execution process of this embodiment is as follows:

[0140] First, calculate the gradient magnitude:

[0141] ;

[0142] in, It is a grayscale image after highlight removal. Calculate the first-order partial derivatives of the grayscale image G in the horizontal and vertical directions respectively.

[0143] Secondly, threshold segmentation can be used to extract pattern lines / boundary regions, forming a pattern mask. :

[0144] ;

[0145] Where t is the threshold, which is usually set empirically.

[0146] right The skeleton Sk=S(Mp) is obtained by refining the image. Based on this, the coarse lines in the image are transformed into a centerline network, which makes it easier to calculate curvature, direction and topology.

[0147] First of all (x)=1, representing the pattern lines and the foreground area; (x)=0 represents the background. Define the skeleton output. For any pixel x1, define its 8 neighboring pixels as x2, ..., x9, arranged clockwise as follows: x2 top, x3 top right, x4 right, x5 bottom right, x6 bottom, x7 bottom left, x8 left, x9 top left. Each pixel takes the value x. i This indicates whether the neighboring pixel is a foreground pixel.

[0148] Define neighborhood foreground number This is used to measure the density of the foreground surrounding the pixel. Let the sequence (x2, x3, x4, x5, x6, x7, x8, x9, x2) be connected in a ring, then:

[0149] ;

[0150] in, This is an indicator function; it returns 1 if the condition is true, and 0 otherwise. Let the initial value be Sk = Mp. For all foreground pixels satisfying Sk(x1) = 1, if the neighborhood number constraint is also satisfied... Connectivity constraints and directional constraints If x1 is not deleted, add it to the deletion set and delete it after the traversal is complete. If no pixel is deleted in a certain traversal, stop the traversal and output Sk at this time.

[0151] In one embodiment, in step S8, the structured feature F is composed of the following sub-feature vectors concatenated:

[0152] Color characteristics It is a color histogram of the standard pattern image;

[0153] Curve characteristics It is a histogram of the orientation angles of each pixel on the geometric skeleton;

[0154] Periodic characteristics It is obtained by calculating the autocorrelation function of the one-dimensional projection signal along the pattern border direction and extracting the period length L corresponding to the main peak position;

[0155] Texture features It is a local binary pattern histogram of a standard pattern image.

[0156] The specific execution process of this embodiment is as follows:

[0157] One-dimensional signal is obtained by sampling along the edge direction. Its autocorrelation function is:

[0158] ;

[0159] in, τ represents the intensity / texture projection signal along the edge direction, and τ is the hysteresis time.

[0160] The most significant period can be estimated as:

[0161] ;

[0162] The corresponding repetition period length can then be obtained:

[0163] ;

[0164] Where L is the actual period length, It is the sampling interval.

[0165] Finally, construct the final feature vector:

[0166] ;

[0167] in, It is a color characteristic, with Hist is a histogram.

[0168] The orientation angle of the skeleton point is calculated as follows:

[0169] ;

[0170] ;

[0171] Periodic characteristics are Texture features are .

[0172] This transforms images into structured, searchable codes that can be used for classification, similarity retrieval, and archiving statistics.

[0173] The above embodiments of the present invention can realize edge period detection and pattern similarity retrieval by constructing a database.

[0174] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0175] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting structured features of glaze patterns based on polarization difference and band fusion, characterized in that, Includes the following steps: S1: Obtain a set of original images of the same glaze pattern area under multiple imaging conditions; the multiple imaging conditions include at least different polarization angle sequences and different illumination band sequences; S2: Perform image registration on the original image set so that the same physical point in all images is mapped to the same pixel coordinate, and obtain the registered image set; S3: Based on the images with different polarization angles in the registered image set, calculate the polarization parameters of each pixel; S4: Generate a binary mask that identifies the specular highlight area based on the polarization parameters and pixel intensity; S5: Based on the polarization parameters and the binary mask, estimate and remove the specular reflection component from the registered image set to obtain a de-spectral textured image; S6: Perform color channel correction on the de-highlighted texture image and output a standard texture image after color and scale normalization; S7: Perform pattern region segmentation and skeletonization processing on the standard pattern image to extract the geometric skeleton of the pattern; S8: Based on the standard pattern image and the geometric skeleton, extract and fuse color, curve, period and texture features to generate the structured features of the glaze pattern.

2. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 1, characterized in that, In step S1, the different polarization angle sequence includes images with polarization directions of 0°, 45°, 90° and 135°; the different illumination band sequence includes images acquired under illumination from at least two different wavelength light sources.

3. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 1, characterized in that, Step S2 specifically includes: S21. Select one frame from the original image set as a reference frame; S22. Extract feature points between the remaining frames and the reference frame and match them to obtain a set of matching point pairs; S23. Based on the set of matching point pairs, solve for the parameters of the affine transformation model using a robust estimation algorithm; S24. Using the affine transformation model parameters, transform the remaining frames of images to the coordinate system of the reference frame to complete the registration.

4. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 1, characterized in that, In step S3, the polarization parameters include Stokes parameters. , The formula for calculating the Stokes parameter is as follows: ; ； ； in, , , , These represent the intensity values ​​at pixel x of the registered images with polarization angles of 0°, 45°, 90°, and 135°, respectively.

5. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 4, characterized in that, The linear polarization degree DoLP and polarization angle AoP of each pixel are calculated based on the Stokes parameters, using the following formula: ； ； in It is the tangent of the four quadrants. It refers to the degree of polarization of the pixel. It is the main polarization direction of the pixel. It is a very small positive number.

6. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 4, characterized in that, Step S4 specifically includes: Calculate the local brightness adaptive threshold for pixel x: ； in, They are respectively based on The mean and standard deviation of the brightness within a local window centered on the α value are given by α, which is a preset empirical coefficient. Define specular candidate mask for: ； ； Where L(x) is the intensity of pixel x, This is the preset polarization threshold.

7. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 1, characterized in that, Step S5 specifically includes: S51. Select the image with the lowest intensity of each pixel from the registered image set. As a base image; S52. Estimate the specular reflection component based on the linear polarization degree DoLP(x) and intensity L(x): ； Where β is a preset weighting coefficient; S53. Within the area marked as a highlight by the binary mask, from the base image... Subtract the specular reflection component from the middle The results are then truncated to obtain a de-highlight texture image. : ； in, It is a truncation function.

8. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 1, characterized in that, Step S6 specifically includes: S61: Regarding the dehighlight texture image For each color channel c∈{R, G, B}, calculate its average intensity outside the highlight region. ; S62: Calculate the gain coefficient for each channel: ； in, The target grayscale mean is... It is a very small positive number; S63: Convert the pixel values ​​of each channel Multiply by the corresponding gain coefficient to obtain the color-normalized image. : 。 9. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 1, characterized in that, Step S7 specifically includes: S71. Calculate the gradient magnitude of the standard pattern image, and obtain a binary mask of the pattern region through threshold segmentation. ; S72, the binary mask A thinning algorithm is applied to iteratively delete boundary pixels that meet the preset deletion conditions until a single-pixel-wide pattern skeleton image Sk is obtained.

10. The method for extracting structured features of glaze patterns based on polarization difference and band fusion according to claim 1, characterized in that, In step S8, the structured feature F is composed of the following sub-feature vectors concatenated: Color characteristics , which is the color histogram of the standard pattern image; Curve characteristics It is a histogram of the orientation angles of each pixel on the geometric skeleton; Periodic characteristics It is obtained by calculating the autocorrelation function of the one-dimensional projection signal along the pattern border direction and extracting the period length L corresponding to the main peak position; Texture features It is a local binary pattern histogram of the standard pattern image.

Images