Tire x-ray image oriented texture primitive extraction method

Abstract

This invention relates to a texture primitive extraction method for tire X-ray images, addressing the challenge of accurately obtaining pixel-by-pixel texture information at the individual cord level while suppressing background interference. It falls under the field of computer vision and image processing technology. The method combines frequency domain analysis to obtain texture direction and spacing, utilizes this information to construct a mesh mask, and further combines background point extraction with real background reconstruction to filter out specific background regions in the original image. Under directional constraints, the remaining texture mesh is continuously tracked to obtain pixel-by-pixel texture information, thus achieving texture primitive extraction. This provides a more reliable foundation for subsequent pathological detection, structural analysis, and cord-level texture modeling.

Description

Technical Field

[0001] This application relates to a method for extracting texture primitives from tire X-ray images, belonging to the field of computer vision and image processing technology. Background Technology

[0002] Tire X-ray inspection is a crucial step in tire non-destructive testing. Its value lies not only in identifying internal tire defects or abnormalities but also in providing a basis for subsequent quality assessment, structural analysis, and safety determination. For tire X-ray images, the truly meaningful object of analysis is not simply grayscale changes, but rather the cord-level structure that constitutes the tire's internal texture network. Texture primitives refer to individual cords within the tire's texture network, representing the smallest structural unit capable of characterizing the network's directionality, periodicity, and connectivity. Since the distribution, continuity, and spacing variations of individual cords directly reflect the tire's internal structural characteristics, the stable and precise extraction of texture information at the individual cord level is of great significance for subsequent structural analysis. Furthermore, tire X-ray images typically exhibit complex anisotropy and multi-layered texture superposition, making the precise extraction of cord-level structures more challenging than general industrial image processing.

[0003] Tire X-ray images typically suffer from low contrast, high noise, significant background grayscale fluctuations, dense and interwoven textures, and easy aliasing of local structures. Public research indicates that the cord regions in tire X-ray images often exhibit overlapping, crossing, and poor transmission effects, with overall local brightness being darker. This directly affects the distinguishability of individual cord boundaries and further weakens the grayscale difference between the background and texture. For such images, the background is not ideally uniform but rather exhibits imaging unevenness and complex structural superposition; simultaneously, the internal cords of the tire often display obvious directionality, periodicity, and continuity, forming a regular but dense texture network structure. These characteristics mean that without specialized processing addressing direction, spacing, and background constraints, individual cords are often difficult to distinguish stably.

[0004] Existing methods for processing tire X-ray images or similar industrial X-ray images mostly focus on determining the presence of anomalies at the overall image, regional, or feature map level, answering "where might the problem be," but often failing to directly provide cord-level results that stably correspond to industry-standard quantitative criteria. A few studies on tire X-ray periodic textures have begun to transform defect detection into tasks such as cord pixel spacing calculation, background connectivity analysis, joint location localization, and quantification. This indicates that what is truly needed in practical engineering is not just coarse-grained anomaly indications, but rather structural quantitative information that can directly serve standardized judgments.

[0005] However, tire X-ray images typically exhibit low contrast, high noise, significant background grayscale fluctuations, dense and interwoven textures, and overlapping local structures. In particular, the cord region may also suffer from poor transmission and localized darkening. For these images, traditional thresholding, conventional filtering, or methods based on local grayscale statistics, while capable of suppressing the background or enhancing edges to some extent, often struggle to simultaneously achieve background suppression and preservation of individual cords. In low signal-to-noise ratio regions, detail breaks, missed detections, and misjudgments are more likely to occur, making the extraction of cord-level structures from complex texture networks still not stable enough.

[0006] On the other hand, while deep learning methods have achieved good results in tire defect detection in recent years, their outputs typically rely on large amounts of labeled data and implicit features learned by the network. The final results are often manifested as classification labels, bounding boxes, or heatmaps, rather than structural parameters that can directly align with quantitative standards such as cord spacing, connectivity, and continuity. To understand the model's underlying principles, related studies often require additional interpretability methods such as Grad-CAM, LRP, CAM, and LIME to visualize the network's regions of interest. This indicates that deep learning solutions still suffer from insufficient interpretability and auditability in engineering applications. Furthermore, such post-hoc interpretation results are not equivalent to consistently outputting cord-level quantitative information that meets quality inspection criteria.

[0007] Furthermore, the cords in tire X-ray images are not randomly distributed but exhibit distinct main directions and spacing characteristics. Using only general enhancement, segmentation, or detection methods often fails to fully utilize these directional and periodic priors, easily leading to discontinuous cord extraction, numerous local misjudgments, and unstable identification of weak anomalies. Especially in complex textured mesh scenarios, accurately obtaining pixel-by-pixel texture information at the level of individual cords while suppressing background interference remains a problem that current technologies need to further address. Summary of the Invention

[0008] To address the problem of accurately obtaining pixel-by-pixel texture information at the level of a single cord while suppressing background interference, this application provides a method for extracting texture primitives from tire X-ray images.

[0009] This application discloses a method for extracting texture primitives from tire X-ray images, comprising:

[0010] S1. Obtain the background components in the original X-ray image of the tire;

[0011] S2. Obtain the texture spacing parameter and texture direction parameter in the original X-ray image of the tire;

[0012] S3. Construct a mesh mask based on the texture spacing parameters and texture direction parameters, and use the mesh mask to extract the coordinates of background points;

[0013] S4. Based on the coordinates of the background points, the background components are corrected to reconstruct a real background image;

[0014] S5. Using the real background image as a segmentation threshold benchmark, compare it pixel by pixel with the original tire X-ray image, delete the background pixel region in the original tire X-ray image, and obtain the texture network candidate region.

[0015] S6. Using the texture direction parameter as a constraint, perform texture tracking in the candidate region of the texture network to obtain pixel-by-pixel texture information and realize texture primitive extraction.

[0016] Preferably, S4 includes:

[0017] The difference between the gray values ​​of each background point in the original tire X-ray image and the gray values ​​of the corresponding positions of the background components is obtained to form discrete residual samples;

[0018] Using the background point coordinates as known constraints and the non-background point region as the region to be assigned values, Laplace diffusion interpolation is performed on the discrete residual sample to obtain a continuous compensation field.

[0019] The continuous compensation field is then superimposed pixel-by-pixel with the background component to reconstruct a true background image.

[0020] Preferably, in step S6, the texture direction parameter is used as a constraint condition to introduce a direction-sensitive operator, and texture tracking is performed in the candidate region of the texture network to obtain pixel-by-pixel texture information, thereby realizing texture primitive extraction.

[0021] First, the directional range from 0° to 180° is discretized into multiple directional layers. On each directional layer, a directional second-order Gaussian derivative kernel with consistent direction is constructed as a direction-sensitive operator.

[0022] Convolution response calculation is performed on candidate regions of the texture network based on the orientation-sensitive operator, and then the orientation enhancement values ​​of each pixel in the candidate regions of the texture network are extracted on different orientation layers.

[0023] The texture direction parameter is used as the main propagation direction. Only the direction layers that are consistent with the main propagation direction or whose corners are inserted into the preset direction tolerance range are retained. The direction enhancement values ​​on different direction layers are superimposed or normalized to form a direction enhancement map for continuous texture tracking.

[0024] First, extract initial seed points along the preset sampling line or local gray-level transition position within the candidate region of the texture network. Then, perform bidirectional extension tracking from each initial seed point to both sides of the main propagation direction.

[0025] In each tracking step, the predicted position for the next step is first constructed based on the current tracking point position and the previous displacement. Then, the candidate pixel with the best value in the direction enhancement map is searched only within a limited search radius near the predicted position, and the candidate pixel is updated as the new tracking point position.

[0026] Preferably, S6 also includes: applying an effective region constraint to the candidate pixel; when the searched candidate pixel falls into an invalid region, is close to the filling boundary, or the direction enhancement response is lower than a preset threshold, terminating the trajectory extension in that direction, so as to avoid the trajectory crossing the background empty area, sliding along the boundary, or entering the non-target texture area.

[0027] Preferably, S2 includes:

[0028] The original X-ray image of the tire is subjected to frequency domain transformation to obtain spectral information. A ring bandpass is used to retain the target frequency band corresponding to the tire texture period in the spectral information.

[0029] Texture direction parameters are extracted from the target frequency band using angle peak finding.

[0030] A wedge filter is constructed based on the peak angle obtained by angle peak finding. The wedge filter is used to perform wedge region inverse transformation on the target frequency band to recover the texture response image in the spatial domain. The texture spacing parameter is extracted from the texture response image by combining the texture direction parameter.

[0031] Preferably, in S3, a mesh mask is constructed based on the texture spacing parameter and texture direction parameter. The mesh mask is used as a binary convolution kernel to perform response calculation on the original X-ray image of the tire. The background point coordinates are extracted by the extreme value order binarization method. The extreme value order binarization method includes: sorting the convolution response results according to the response intensity, and selecting the pixel points that satisfy the background response characteristics as background point coordinates according to the preset extreme value order rules.

[0032] Preferably, the pixel-by-pixel texture information includes at least one of the following: whether each pixel belongs to a texture structure, texture direction attribution information, and texture continuity information.

[0033] The beneficial effects of this application are that it obtains texture direction and texture spacing by combining frequency domain analysis, constructs a mesh mask using the direction and spacing information, further combines background point extraction and real background reconstruction, filters out the determined background area in the original image, and continuously tracks the remaining texture mesh under direction constraints, thereby obtaining pixel-by-pixel texture information and realizing texture primitive extraction, thus providing a more reliable foundation for subsequent pathological detection, structural analysis and curtain-level texture modeling. Attached Figure Description

[0034] Figure 1 This is a flowchart of the method described in this application;

[0035] Figure 2 The diagram shows the frequency domain analysis and texture direction, where (a) is the original image, (b) is the frequency domain analysis result, and (c) is the texture direction extraction diagram.

[0036] Figure 3 The diagram shows the structure of the mesh mask and the background point extraction. (a) is the unidirectional texture spatial domain response map, (b) is the schematic diagram of the mesh mask structure, (c) is the original image, and (d) is the background point extraction map.

[0037] Figure 4 The images show the results of real background reconstruction and background removal. (a) is the real background, (b) is the background obtained directly from frequency domain operations (control), (c) is the background removal result, and (d) is the original image (control).

[0038] Figure 5 The images are pixel texture extraction results, where (a) is the pixel-by-pixel texture extraction result and (b) is the unidirectional extraction result. Detailed Implementation

[0039] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0040] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0041] The present application will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the application.

[0042] This embodiment of the texture primitive extraction method for tire X-ray images first acquires the original tire X-ray image and its initial background components, then extracts texture spacing and texture direction parameters through frequency domain analysis; next, it constructs a mesh mask and extracts background point coordinates, then reconstructs the real background image using Laplacian diffusion interpolation; finally, it filters out background pixels using the real background as a threshold to obtain candidate regions for the texture mesh; and finally, it performs continuous texture tracing under directional constraints, outputting pixel-by-pixel texture information to complete the extraction of texture primitives at the single cord level. Specifically, it includes:

[0043] Step 1: Obtain the raw X-ray image of the tire. Background components ;

[0044] Raw X-ray image of a tire It includes texture (curtain lines) and background (curtain line gaps and non-textured areas). The background component is the slowly varying low-frequency part of the image, which is relatively separated from the high-frequency periodic information of the texture. Background separation methods can be used to initially extract background components, providing a benchmark for subsequent realistic background reconstruction.

[0045] Existing methods for extracting complex image backgrounds, such as large-scale median filtering or morphological closing operations, are employed. In this embodiment, a circular structuring element with a set radius of pixels is used to perform morphological closing operations on the original image, followed by Gaussian filtering for smoothing, to obtain the initial background components. ;

[0046] Step 2: Obtain the raw X-ray image of the tire. Texture spacing parameters and texture direction parameters ;

[0047] The texture of tire cords exhibits obvious periodicity and directionality, appearing as a pair of symmetrical bright spots in the frequency domain. By transforming the image to the frequency domain using a two-dimensional Fourier transform, filtering out non-periodic components with a circular bandpass filter, determining the main direction of the texture through angle peak finding, and recovering the texture response in a single direction using a wedge inverse transform, the texture spacing can be accurately measured in the spatial domain.

[0048] Specifically:

[0049] Frequency domain transformation: transforming the original X-ray image of a tire After applying the Hanning window, a two-dimensional Fast Fourier Transform (FFT) is performed to obtain the complex spectrum. And calculate the amplitude spectrum Move the zero frequency to the center;

[0050] Ring bandpass: Based on the prior range of the cord spacing, calculate the frequency radius and construct a ring bandpass filter. Only the spectral components within the ring band are retained to obtain the target frequency band. .

[0051] Peak finding at angles: targeting frequency bands Transform to polar coordinates, accumulate radial energy for each angle, and find the angle corresponding to the maximum energy. Since the direction of the bright spots in the spectrum is perpendicular to the texture direction, the principal direction of the texture is... .

[0052] Inverse transformation of the wedge region: Construct a wedge filter centered on [the target area]. ,Will Multiply To obtain the texture response image In texture response images Multiple straight lines are sampled along the top edge perpendicular to the texture direction. The autocorrelation function of the sampled grayscale sequence is calculated, and the delay corresponding to the first peak is taken as the period. The median of the results from multiple sampling lines is then taken to obtain the texture spacing parameter. .

[0053] Obtain globally stable texture orientation parameters and texture spacing parameters This provides prior information for subsequent mesh mask construction and orientation constraint tracing.

[0054] Step 3: Construct a mesh mask based on the texture spacing and texture direction parameters, and use the mesh mask to extract the coordinates of background points. ;

[0055] The tire texture mesh is formed by the interweaving of multiple layers of cords, with a clean background in the center of each mesh. Using the texture spacing and direction parameters obtained in step 2, a binarized convolution kernel matching the mesh shape is constructed. This kernel is then applied to the original image to perform a convolution response. The response intensity at the center of each mesh is extremely low; these background points can be automatically located using extremum order binarization.

[0056] A mesh mask is constructed, and its geometric distribution is determined by both texture spacing and texture direction parameters, ensuring that the mesh shape and size of the textured mesh generated by the binarized convolution kernel and the overlapping of multiple textures are approximately the same. The mesh mask is then used as the convolution kernel to process the raw tire X-ray image. A convolution operation is performed to obtain a response map. In the center region of the mesh, the response value is low because the convolution kernel covers the surrounding texture and the center is 0; on texture lines, the response value is higher. All pixels in the response map are sorted in ascending order of response intensity (lower response is more likely to be a background point). Based on a preset extreme value order rule, such as selecting the N pixels with the lowest response, or selecting pixels with responses below the global mean, the coordinates of background points are extracted.

[0057] This step automatically and accurately extracts the coordinates of background points located at the center of the mesh, providing sparse but reliable constraint samples for real background reconstruction.

[0058] Step 4: Based on background point coordinates For background components After correction, a realistic background image is obtained through reconstruction. ;

[0059] Initial background components There is local bias, while the original tire X-ray image at the background point... This represents the real background. The residuals at the background points are calculated. Laplacian diffusion interpolation is used to smoothly propagate the discrete residuals to the entire image region, resulting in a continuous compensation field, which is then superimposed onto the initial background components. By doing so, a continuous image that closely approximates the real background can be obtained; specifically,

[0060] Calculate the discrete residual samples for each background point. The difference between the gray values ​​of each background point in the original tire X-ray image and the corresponding gray values ​​of the background components is obtained to form discrete residual samples. ;

[0061] Perform Laplace diffusion interpolation to determine the coordinates of the background points. Given a set of points, and using non-background point regions as the regions to be assigned values, Laplace diffusion interpolation is performed on the discrete residual samples to obtain a continuous compensation field. pixel-by-pixel overlay At the background point At non-background points, the diffusion result provides a smooth transition.

[0062] This step eliminates local biases in the initial background components, resulting in a continuous, smooth, and accurate image of the true background. This provides a high-fidelity segmentation threshold for subsequent background removal.

[0063] Step 5: Using the real background image as the segmentation threshold, compare it pixel-by-pixel with the original tire X-ray image, delete the background pixel regions in the original tire X-ray image, and obtain the texture mesh candidate regions. ;

[0064] In a real background image, the grayscale values ​​of the background area accurately reflect the background level of the original image. Textures (curtains) are usually darker than the background (low X-ray transmittance), so pixels in the original image with grayscale values ​​higher than or equal to the real background can be identified as background and removed, thus significantly narrowing the texture search range.

[0065] Pixel-by-pixel comparison of the original image With real background image Generate candidate region mask: If the gray value of a pixel in the original image meets the judgment condition of being higher than or equal to the gray value of the corresponding real background, then the pixel is judged as a background pixel and filtered out.

[0066] This step quickly removes most of the background pixels, resulting in candidate regions for the texture mesh. This significantly reduces the computational cost of subsequent texture tracing and avoids background interference.

[0067] Step 6: Use the texture direction parameter as a constraint in the candidate region of the texture mesh. Texture tracing is performed to obtain pixel-by-pixel texture information, thereby enabling texture primitive extraction.

[0068] The tire cords are continuous curves with a known main propagation direction. By constructing a direction-sensitive operator to enhance the texture response along the main propagation direction and performing prediction-search tracking within candidate regions, single cords can be stably extracted, avoiding missed detections of intersections, breaks, or weak texture areas.

[0069] By using texture direction parameters as constraints and introducing a direction-sensitive operator, texture tracing is performed in the candidate regions of the texture mesh to obtain pixel-by-pixel texture information, thereby achieving texture primitive extraction.

[0070] First, the directional range from 0° to 180° is discretized into multiple directional layers, and then a directional second-order Gaussian derivative kernel with the same direction is constructed on each directional layer.

[0071]

[0072] in, , , It is a Gaussian function. For direction;

[0073] The directional second-order Gaussian derivative kernel is used as a direction-sensitive operator;

[0074] Candidate regions of the texture mesh are selected based on the direction-sensitive operator. Perform convolution response calculation, and then extract the orientation enhancement values ​​of each pixel in the candidate region of the texture network at different orientation layers. ;

[0075] Texture direction parameters As the main propagation direction, a preset directional tolerance is set, retaining only directional layers that are consistent with the main propagation direction or whose corners fall within the preset directional tolerance range, and adjusting the directional enhancement values ​​on different directional layers. The overlay or normalization process is performed to form a direction enhancement map for continuous texture tracing;

[0076] First in the candidate region of the texture mesh Initial seed points are extracted from the inner edge of the preset sampling line or local gray-level transition positions. Then, starting from each initial seed point, the samples are distributed to both sides of the main propagation direction. and Perform bidirectional extended tracking;

[0077] In each step of the tracking, first, based on the current tracking point position... and preceding displacement Construct the next predicted location Then only at the predicted location Within a finite search radius, search for the candidate pixel with the best value in the direction enhancement image and update that candidate pixel as the new tracking point location.

[0078] An effective region constraint is applied to the candidate pixels. When the searched candidate pixels fall into an invalid region, are close to the fill boundary, or the direction enhancement response is lower than a preset threshold, the trajectory extension in that direction is terminated to avoid the trajectory crossing the background empty area, sliding along the boundary, or entering the non-target texture area.

[0079] This step enables continuous and accurate extraction of a single cord, maintaining the integrity of the trajectory even in areas with weak or intersecting textures, and outputting a pixel-by-pixel texture marker map.

[0080] Step 7: Output pixel-by-pixel texture information;

[0081] All trajectory points obtained in step 6 are summarized to form a binary labeled map, which can be supplemented with direction attribution and continuity information; specifically, a labeled matrix of the same size as the original image is initialized. For each pixel on each track obtained from the tracking, set... In addition, the local orientation and continuity label of each texture pixel can be recorded. The final output is pixel-by-pixel texture information, which includes at least the labeling information of whether each pixel belongs to the texture structure, and may also include at least one of texture orientation information and texture continuity information.

[0082] This application enables pixel-by-pixel extraction of texture information: compared with methods that only output the binarized result of the texture network or a coarse abnormal response map, this application can obtain pixel-by-pixel texture information in tire X-ray images, which is more conducive to subsequent texture primitive modeling, structural analysis and pathological detection.

[0083] This application can simultaneously achieve background suppression and texture preservation: First, it obtains the compensation information of the original image relative to the background components based on the background points, and uses the compensation information to correct the background components to reconstruct the real background. Then, it filters out the determined background areas, reduces the impact of background grayscale fluctuations on texture extraction, and makes subsequent texture tracking more stable.

[0084] This application can make full use of the prior knowledge of the directionality and periodicity of the texture: This application extracts the texture spacing and texture direction through frequency domain analysis, and constructs mesh mask and direction constraint conditions accordingly, so that the extraction process is more in line with the actual structural distribution of the tire texture mesh.

[0085] This application can improve the extraction accuracy in complex texture scenes: for cross textures, dense textures and local weak texture areas, this application can maintain the continuity of texture structure, reduce local misjudgments, and thus improve the accuracy of pathological detection based on texture structure.

[0086] The method proposed in this application has good engineering application value: This application is aimed at the typical industrial scenario of tire X-ray images, and can provide reliable underlying texture information support for subsequent pathological identification, anomaly localization and internal structure analysis.

[0087] While this application has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of this application. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of this application as defined by the appended claims. It should be understood that different dependent claims and features herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other embodiments.

Claims

1. A method for extracting texture primitives from tire X-ray images, characterized in that, include: S1. Obtain the background components in the original X-ray image of the tire; S2. Obtain the texture spacing parameter and texture direction parameter in the original X-ray image of the tire; S3. Construct a mesh mask based on the texture spacing parameters and texture direction parameters, and use the mesh mask to extract the coordinates of background points; S4. Based on the coordinates of the background points, the background components are corrected to reconstruct a real background image; S5. Using the real background image as a segmentation threshold benchmark, compare it pixel by pixel with the original tire X-ray image, delete the background pixel region in the original tire X-ray image, and obtain the texture network candidate region. S6. Using the texture direction parameter as a constraint, perform texture tracking in the candidate region of the texture network to obtain pixel-by-pixel texture information and realize texture primitive extraction.

2. The tire-oriented X-ray image texture primitive extraction method according to claim 1, wherein S4 include: The difference between the gray values ​​of each background point in the original tire X-ray image and the gray values ​​of the corresponding positions of the background components is obtained to form discrete residual samples; Using the background point coordinates as known constraints and the non-background point region as the region to be assigned values, Laplace diffusion interpolation is performed on the discrete residual sample to obtain a continuous compensation field. The continuous compensation field is then superimposed pixel-by-pixel with the background component to reconstruct a true background image.

3. The method for extracting texture primitives from tire X-ray images according to claim 1, characterized in that, In step S6, the texture direction parameter is used as a constraint condition, and a direction-sensitive operator is introduced to perform texture tracking in the candidate region of the texture network to obtain pixel-by-pixel texture information and achieve texture primitive extraction. First, the directional range from 0° to 180° is discretized into multiple directional layers. On each directional layer, a directional second-order Gaussian derivative kernel with consistent direction is constructed as a direction-sensitive operator. Convolution response calculation is performed on candidate regions of the texture network based on the orientation-sensitive operator, and then the orientation enhancement values ​​of each pixel in the candidate regions of the texture network are extracted on different orientation layers. The texture direction parameter is used as the main propagation direction. Only the direction layers that are consistent with the main propagation direction or whose corners are inserted into the preset direction tolerance range are retained. The direction enhancement values ​​on different direction layers are superimposed or normalized to form a direction enhancement map for continuous texture tracking. First, extract initial seed points along the preset sampling line or local gray-level transition position within the candidate region of the texture network. Then, perform bidirectional extension tracking from each initial seed point to both sides of the main propagation direction. In each tracking step, the predicted position for the next step is first constructed based on the current tracking point position and the previous displacement. Then, the candidate pixel with the best value in the direction enhancement map is searched only within a limited search radius near the predicted position, and the candidate pixel is updated as the new tracking point position.

4. The method for extracting texture primitives from tire X-ray images according to claim 3, characterized in that, S6 further includes: applying an effective region constraint to the candidate pixel; when the searched candidate pixel falls into an invalid region, is close to the filling boundary, or the direction enhancement response is lower than a preset threshold, terminating the trajectory extension in that direction, so as to avoid the trajectory crossing the background empty area, sliding along the boundary, or entering the non-target texture area.

5. The tire X-ray image oriented texture primitive extraction method of claim 1, wherein, S2 includes: The original X-ray image of the tire is subjected to frequency domain transformation to obtain spectral information. A ring bandpass is used to retain the target frequency band corresponding to the tire texture period in the spectral information. Texture direction parameters are extracted from the target frequency band using angle peak finding. A wedge filter is constructed based on the peak angle obtained by angle peak finding. The wedge filter is used to perform wedge region inverse transformation on the target frequency band to recover the texture response image in the spatial domain. The texture spacing parameter is extracted from the texture response image by combining the texture direction parameter.

6. The method for extracting texture primitives from tire X-ray images according to claim 1, characterized in that, In step S3, a mesh mask is constructed based on the texture spacing parameter and texture direction parameter. The mesh mask is used as a binary convolution kernel to perform response calculation on the original tire X-ray image. The background point coordinates are extracted by the extreme value order binarization method. The extreme value order binarization method includes: sorting the convolution response results according to the response intensity, and selecting the pixel points that satisfy the background response characteristics as background point coordinates according to the preset extreme value order rules.

7. The tire X-ray image oriented texture primitive extraction method of claim 1, wherein, The pixel-by-pixel texture information includes at least one of the following: whether each pixel belongs to a texture structure, texture direction attribution information, and texture continuity information.

8. A computer-readable storage device storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the texture primitive extraction method for tire X-ray images as described in any one of claims 1 to 7.

9. A texture primitive extraction apparatus for tire X-ray images, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the texture primitive extraction method for tire X-ray images as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the texture primitive extraction method for tire X-ray images as described in any one of claims 1 to 7.

Images