An image quantification evaluation system for the abutment relationship of an oral prosthesis

Abstract

The application discloses a kind of oral prosthetic body abutment relationship image quantitative evaluation systems, belong to oral prosthetic body image processing technical field, including image acquisition and pretreatment module, feature atlas construction module, topological analysis calculation module and quantitative evaluation mapping module.Image acquisition and pretreatment module obtains the three-dimensional scanning image of prosthetic body in situ state, obtains two-dimensional development image by gray mapping and two-dimensional development, separates out highlight contact zone by multi-scale morphological opening and closing reconstruction operation;Feature atlas construction module constructs morphological feature atlas according to the morphological gradient distribution characteristics of highlight contact zone under different scales;Topological analysis calculation module tracks the continuous boundary of abutment contact zone in morphological feature atlas, and calculates the length, width and area quantitative value of contact zone;Quantitative evaluation mapping module maps quantitative value to prosthetic three-dimensional space frequency domain, generates quantitative index for evaluating the closeness of abutment relationship.

Description

Technical Field

[0001] This invention relates to the field of dental prosthesis image processing technology, specifically to a dental prosthesis adjacency relationship image quantitative evaluation system. Background Technology

[0002] After dental prostheses are placed, the tightness of the contact between adjacent prostheses directly affects food impaction, periodontal health, and patient comfort. Current clinical assessment methods primarily rely on the resistance felt when flossing and visual interpretation of bite paper imprints. These methods are highly dependent on the operator's subjective experience, providing only qualitative results such as looseness or tightness, lacking objective and unified quantitative standards. With the widespread use of intraoral scanning equipment, some digital systems can acquire three-dimensional models of the prosthesis in place and perform two-dimensional unfolding of the adjacent areas in the software. Contact zones are segmented using simple grayscale thresholding, and then the projected area or linear dimensions of the contact zone are measured. While this method achieves preliminary quantification, single threshold segmentation is unstable when dealing with grayscale gradients, artifacts at the prosthesis margins, and subtle surface undulations. The contact zone boundaries are blurred, and noise interference is significant. Furthermore, fixed processing parameters cannot be adaptively adjusted according to the actual size of the contact zone, resulting in insufficient adaptability to individual differences in different prostheses. Furthermore, using only geometric measurements from two-dimensional unfolded diagrams as the basis for evaluation ignores the actual curvature of the proximal surfaces of the prosthesis in three-dimensional space and the frequency domain characteristics of the contact distribution. This results in a significant deviation between the obtained indicators and the clinically perceived tightness. Deformation during the two-dimensional unfolding process also introduces measurement errors, causing parameters such as the contact area and length to fail to accurately reflect the compactness of the curved surface contact. Therefore, how to accurately separate the contact zone region from a multi-scale morphological perspective and map the quantitative information to the three-dimensional frequency domain to generate an indicator that comprehensively reflects the tightness of the contact is a problem that needs to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide a quantitative evaluation system for the adjacency relationship of dental prostheses, in order to solve the problems of strong subjectivity and the inability of two-dimensional geometric measurements to accurately reflect the contact characteristics of three-dimensional curved surfaces in existing adjacency relationship evaluation technologies.

[0004] The objective of this invention can be achieved through the following technical solutions:

[0005] This invention provides a quantitative evaluation system for the adjacency relationship of dental prostheses, which includes an image acquisition and preprocessing module, a feature map construction module, a topology analysis and calculation module, and a quantitative evaluation and mapping module.

[0006] The image acquisition and preprocessing module acquires a 3D scan image of the dental prosthesis in its in-place state. It then performs grayscale mapping and 2D unfolding on the 3D scan image to obtain a 2D unfolded image of the prosthesis's adjacent regions. The module further performs multi-scale morphological decomposition on the 2D unfolded image, extracting structural elements of the adjacent relationship image at each scale, and separating the highlight contact zone region through morphological opening and closing reconstruction.

[0007] As a technical solution of the present invention, after obtaining the two-dimensional unfolded image and before performing multi-scale morphological decomposition, a preprocessing operation is also included for the two-dimensional unfolded image. Specifically, anisotropic diffusion filtering is applied to the two-dimensional unfolded image to remove image noise and retain edge information, resulting in a filtered image; adaptive histogram equalization is performed on the filtered image to enhance the contrast between adjacent contact zones and background regions, resulting in a contrast-enhanced image; the edge contour of the restoration is identified in the contrast-enhanced image, and the pixel grayscale values ​​of the regions outside the edge contour of the restoration are set to zero, resulting in a two-dimensional unfolded image to be decomposed after removing background interference. By preprocessing the two-dimensional unfolded image, noise and background interference can be effectively eliminated, image quality can be improved, and a foundation can be laid for the accurate separation of the subsequent bright contact zone region.

[0008] Preferably, the specific steps for separating the bright contact zone region include: performing a grayscale morphological erosion operation on the two-dimensional unfolded image, using circular structuring elements of different sizes to sequentially erode the image, obtaining eroded images at multiple scales; performing a grayscale morphological dilation operation on the eroded image at each scale, using circular structuring elements of the same size as the erosion operation to dilate, obtaining morphological opening operation results at multiple scales; subtracting the morphological opening operation results at each scale from the two-dimensional unfolded image to obtain candidate regions of bright contact zones at multiple scales; and performing a morphological closing reconstruction operation on the candidate regions of bright contact zones to fill the internal holes of the candidate regions, obtaining bright contact zone regions with enhanced continuity. This multi-scale decomposition combined with morphological operations can effectively suppress useless texture interference in the image and accurately separate continuous band-shaped regions representing the degree of close contact between adjacent areas.

[0009] The feature map construction module constructs morphological feature maps of adjacent contact zones based on the morphological gradient distribution characteristics of the highlighted contact zone region at different scales. Specifically, it calculates the morphological gradient of the highlighted contact zone region at each scale, where the morphological gradient is the pixel difference between the dilation and erosion results of the highlighted contact zone region. The morphological gradients at all scales are normalized, and the normalized gradient values ​​are stacked sequentially in ascending order of scale to form a three-dimensional gradient distribution tensor. Along the direction of increasing scale, the maximum gradient response value at each pixel position is extracted from the three-dimensional gradient distribution tensor to obtain a maximum gradient response map. The maximum gradient response map is then fused pixel-level with the two-dimensional unfolded image to generate the morphological feature map. Preferably, when stacking the normalized gradient values ​​sequentially in ascending order of scale, cubic spline interpolation is used to smooth the gradient values ​​in the scale direction. By capturing the gradient evolution characteristics of the contact zone structure and extracting the maximum response in multi-scale space, the generated morphological feature map can stably highlight the true boundary of the contact zone, providing a high-quality image foundation for subsequent accurate quantization calculations.

[0010] The topology analysis and calculation module tracks the continuous boundaries of adjacent contact zones in the morphological feature map and uses topological structure analysis methods to calculate the length, width, and area quantization values ​​of the contact zones. As a preferred embodiment of this invention, the specific implementation is as follows: a grayscale threshold is set in the morphological feature map; pixels larger than the grayscale threshold are marked as seed points for the contact zones; eight-neighbor connected components are marked for each seed point to obtain multiple initial contact zone connected components; the equivalent ellipse major axis length and equivalent ellipse minor axis length of each initial contact zone connected component are calculated; the equivalent ellipse major axis length is used as the initial length of the contact zone, and the equivalent ellipse minor axis length is used as the initial width of the contact zone; a skeletonization extraction operation is performed on each initial contact zone connected component to obtain a central skeleton line with a single pixel width; the total pixel length of the central skeleton line is calculated as the precise length of the contact zone; the total number of pixels contained in each initial contact zone connected component is counted, and the precise area of ​​the contact zone is calculated by combining the spatial resolution parameters of the two-dimensional unfolded image. By combining connected component analysis, skeletonization extraction, and spatial resolution, this module can accurately extract geometric dimension information reflecting the degree of adjacent contact from morphological feature maps.

[0011] The quantitative assessment mapping module maps the quantified values ​​of the contact band's length, width, and area to the three-dimensional spatial frequency domain of the restoration, generating a quantitative index for assessing the tightness of the adjacency relationship. The specific mapping process includes: acquiring the three-dimensional surface coordinates of the adjacent region in the three-dimensional scan image; projecting the precise length, width, and area of ​​the contact band onto the three-dimensional surface coordinates to obtain the three-dimensional contact band length, width, and area; performing a Fourier transform on the three-dimensional contact band length, width, and area to extract the low-frequency component amplitude values ​​in the frequency domain; and weighting and summing the low-frequency component amplitude values ​​according to preset weighting coefficients to generate a comprehensive adjacency tightness index. Preferably, the preset weighting coefficients have an area weight of 0.5, a length weight of 0.3, and a width weight of 0.2. This mapping module restores the geometric parameters measured from the two-dimensional image to the three-dimensional real space and converts them to frequency domain analysis, eliminating the influence of surface deformation on contact assessment and comprehensively characterizing the overall tightness of the contact band.

[0012] Preferably, the specific steps for projecting the precise length, width, and area of ​​the contact band onto a three-dimensional surface coordinate system include: marking the three-dimensional point cloud corresponding to the pixel positions of the two-dimensional unfolded image in the three-dimensional scan image of the prosthesis to obtain a set of three-dimensional point cloud coordinates; calculating the Euclidean distance between the three-dimensional point cloud coordinates corresponding to each pixel on the central skeleton line, and summing all Euclidean distances to obtain the length of the three-dimensional contact band; calculating the pixel width perpendicular to the central skeleton line within the initial contact band's connected domain, calculating the three-dimensional distance along the normal direction based on the three-dimensional point cloud coordinates corresponding to the pixel width, and taking the average of all normal direction three-dimensional distances as the width of the three-dimensional contact band; and calculating the area of ​​the surface patch enclosed by the three-dimensional point clouds corresponding to all pixels within the initial contact band's connected domain, and taking the area of ​​the surface patch as the area of ​​the three-dimensional contact band. The three-dimensional point cloud coordinates corresponding to each pixel on the central skeleton line are obtained through a nearest neighbor matching algorithm. Through the above three-dimensional projection steps, the precise positioning and reconstruction of the two-dimensional measurements on the real anatomical surface is achieved.

[0013] As a further optimization of this invention, before mapping the quantized values ​​of the length, width, and area of ​​the contact strip to the three-dimensional spatial frequency domain of the restoration, a consistency verification operation is performed on the quantized values. The specific steps are as follows: calculate the standard deviation of the pixel grayscale values ​​within the initial connected domain of the contact strip, and compare the standard deviation with a preset fluctuation threshold; when the standard deviation is less than the preset fluctuation threshold, the quantized values ​​are determined to be consistent, and the process directly proceeds to the three-dimensional spatial frequency domain mapping step; when the standard deviation is greater than or equal to the preset fluctuation threshold, the quantized values ​​are determined to have local anomalies, and the initial connected domain of the contact strip is segmented before recalculating the length, width, and area quantized values. This consistency verification effectively avoids calculation deviations caused by uneven grayscale within the contact strip, ensuring the accuracy and reliability of the geometric parameters input to the frequency domain mapping.

[0014] Preferably, when a local anomaly is determined in the quantization value, the specific steps for segmenting and recalculating the initial contact zone connected domain include: detecting local minimum grayscale points in the initial contact zone connected domain; using these local minimum grayscale points as segmentation seed points; executing a watershed segmentation algorithm to divide the initial contact zone connected domain into multiple sub-connected domains; calculating the length, width, and area quantization values ​​of each sub-connected domain to obtain multiple sets of sub-quantization values; calculating the mean and variance of the multiple sets of sub-quantization values; selecting the sub-connected domain corresponding to the sub-quantization value with the smallest variance as the main contact zone region; and replacing the original contact zone length, width, and area quantization values ​​with the sub-quantization values ​​of the main contact zone region. This segmentation and recalculation strategy can autonomously select the most stable contact sub-regions, further improving the robustness and accuracy of the quantization assessment.

[0015] As another technical solution of the present invention, after generating the comprehensive adjacency density index, the method further includes classifying the index into levels. Specifically, the comprehensive adjacency density index of multiple standard dental prostheses is obtained, and the multiple standard dental prostheses are clinically rated to obtain clinical rating labels. Using the comprehensive adjacency density index as input features and the clinical rating labels as output targets, a support vector machine classifier is trained, with the kernel function of the support vector machine classifier using radial basis functions. The comprehensive adjacency density index calculated for the prosthesis to be evaluated is input into the trained support vector machine classifier, and the adjacency density level of the prosthesis to be evaluated is output. By mapping the quantitative index to an intuitive clinical level, the evaluation results are more easily applied by clinicians to the judgment and adjustment of prosthesis fit.

[0016] The beneficial effects of this invention are:

[0017] Multi-scale morphological decomposition is performed on the 2D unfolded image. Circular structuring elements of different sizes are sequentially eroded and dilated to perform morphological opening operations. The results of each scale opening operation are subtracted from the original image to obtain candidate regions for bright contact zones. Then, morphological closing reconstruction is used to fill the voids within these candidate regions, resulting in bright contact zone regions with enhanced continuity. Based on this, the morphological gradients of the bright contact zone regions at each scale are calculated. After normalizing all scale gradients, they are stacked into a 3D gradient distribution tensor. The maximum gradient response value at each pixel location is extracted along the scale-increasing direction, generating a maximum gradient response map, which is then fused with the 2D unfolded image to form a morphological feature map. The combination of multi-scale structuring elements can adapt to contact zones of different sizes. The opening and closing reconstruction process suppresses noise while preserving the true edge positions of the contact zones. Morphological gradient stacking and maximum response extraction highlight the contact zone boundary structure from the gradient distribution, making the tracked continuous boundaries of the contact zones more accurate, thereby improving the accuracy of subsequent length, width, and area quantization values.

[0018] The quantified values ​​of the contact zone length, width, and area obtained from morphological feature mapping are projected onto the 3D curved surface point cloud of the restoration through nearest neighbor matching. The Euclidean distances between the 3D point clouds corresponding to the central skeleton line are calculated and accumulated to obtain the 3D contact zone length. The average 3D distance along the normal direction is calculated to obtain the 3D contact zone width. The area of ​​the curved surface area enclosed by the point clouds within the contact zone region is calculated to obtain the 3D contact zone area. Fourier transforms are performed on the 3D contact zone length, 3D contact zone width, and 3D contact zone area to extract the low-frequency component amplitude values ​​in the frequency domain. The values ​​are then weighted and summed according to an area weight of 0.5, a length weight of 0.3, and a width weight of 0.2 to generate a comprehensive adjacency density index. Three-dimensional surface projection allows contact parameters to reflect the actual curvature of adjacent surfaces, avoiding geometric distortion caused by two-dimensional unfolding; frequency domain transformation converts the spatial distribution of contact into a frequency domain expression, with low-frequency components reflecting the overall tightness of the contact zone and filtering out high-frequency fluctuation interference caused by local micro-gaps; weighted fusion assigns a higher proportion to the area, which is consistent with the clinical perception that changes in contact area are more sensitive, making the generated comprehensive tightness index more consistent with the clinical judgment of the tightness of the adjacent relationship. Attached Figure Description

[0019] The invention will now be further described with reference to the accompanying drawings.

[0020] Figure 1 This is a schematic diagram of the module structure of the image-based quantitative assessment system for the adjacency relationship of dental prostheses;

[0021] Figure 2 This is a flowchart of multi-scale morphological decomposition and contact zone extraction of two-dimensional unfolded images of adjacent areas of the restoration;

[0022] Figure 3 This is a flowchart of the morphological feature map construction process for adjacent contact zones;

[0023] Figure 4 This is a flowchart for the quantitative analysis of contact strip length, width, and area; Detailed Implementation

[0024] 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.

[0025] See Figure 1This invention provides a quantitative evaluation system for the adjacency relationship of oral prostheses, comprising: an image acquisition and preprocessing module, which acquires a three-dimensional scan image of the oral prosthesis in its in-situ state, performs grayscale mapping and two-dimensional unfolding processing on the three-dimensional scan image to obtain a two-dimensional unfolded image of the adjacency region of the prosthesis; the image acquisition and preprocessing module performs multi-scale morphological decomposition on the two-dimensional unfolded image, extracts the structural elements of the adjacency relationship image at each scale, and separates the bright contact zone region through morphological opening and closing reconstruction operation; a feature map construction module, which constructs a morphological feature map of the adjacency contact zone based on the morphological gradient distribution characteristics of the bright contact zone region at different scales; a topology analysis calculation module, which tracks the continuous boundary of the adjacency contact zone in the morphological feature map and calculates the length, width, and area quantification values ​​of the contact zone using topological structure analysis methods; and a quantitative evaluation mapping module, which maps the length, width, and area quantification values ​​of the contact zone to the three-dimensional spatial frequency domain of the prosthesis to generate a quantitative index for evaluating the tightness of the adjacency relationship.

[0026] In specific implementation, please refer to Figure 2 After obtaining the two-dimensional unfolded image of the adjacent region of the restoration, preprocessing operations are performed on the unfolded image. Anisotropic diffusion filtering is applied to remove image noise and preserve edge information. Anisotropic diffusion filtering is achieved by solving partial differential equations. The diffusion coefficient is adaptively adjusted according to the local gradient of the image. The diffusion coefficient is larger in flat regions with small gradients to smooth noise, and the diffusion coefficient approaches zero in edge regions with large gradients to preserve edges. In the specific implementation, the number of diffusion iterations is set to 10, the time step is 0.2, the thermal conductivity is 1.0, and the edge threshold parameter in the diffusion coefficient calculation is set to 15.0. The two-dimensional unfolded image is used as input, and after the above iterative diffusion processing, the filtered image is output.

[0027] Next, adaptive histogram equalization is performed on the filtered image to enhance the contrast between adjacent contact zones and the background region. Adaptive histogram equalization divides the filtered image into locally overlapping blocks of size 8 pixels by 8 pixels. A gray-level histogram is independently calculated for each local block, and the histogram is cropped based on a contrast limiting parameter of 0.01. Pixels exceeding the limit are evenly distributed across the gray levels of the histogram, and histogram equalization is then performed. For each equalized local block, bilinear interpolation is used to eliminate boundary artifacts between blocks, resulting in a contrast-enhanced image.

[0028] In contrast-enhanced images, the edge contours of the restoration are identified, and the gray values ​​of pixels outside the restoration's edge contours are set to zero to obtain a two-dimensional unfolded image to be decomposed, free from background interference. The Canny edge detection method is used for restoration edge contour identification. First, the contrast-enhanced image is smoothed using a Gaussian kernel with a standard deviation of 1.2. The image gradient magnitude and direction are calculated, and non-maximum suppression is performed. Dual-threshold detection and edge connection are then performed with a low threshold of 30 and a high threshold of 90 to obtain the restoration's edge contours. Using the restoration's edge contours as boundaries, the regions inside and outside the contours are marked. A flood fill algorithm is used to set the gray values ​​of all pixels outside the contours to zero, while the gray values ​​of pixels inside the contours remain unchanged, resulting in the two-dimensional unfolded image to be decomposed.

[0029] After obtaining the 2D unfolded image to be decomposed, multi-scale morphological decomposition is performed. Gray-scale morphological erosion is then performed, using circular structuring elements of different sizes to erode the unfolded image sequentially. The radius of the selected circular structuring elements ranges from 1 pixel to 15 pixels, with a 1-pixel interval, for a total of 15 scales. For each scale, a circular structuring element of the corresponding radius is used to erode the unfolded image, resulting in the eroded image for that scale.

[0030] A grayscale morphological dilation operation is performed on the eroded image at each scale, using a circular structuring element of the same size as the erosion operation. That is, for each scale, a circular structuring element of the same radius is used to dilate the eroded image, yielding the morphological operation result corresponding to that scale.

[0031] By subtracting the morphological opening results at each scale from the two-dimensional unfolded image to be decomposed, candidate regions of bright contact zones at multiple scales are obtained. The subtraction operation is performed pixel by pixel, subtracting the gray values ​​of corresponding pixels, and setting the results less than zero to zero, thus obtaining candidate regions of bright contact zones.

[0032] A morphological closure reconstruction operation is performed on the candidate regions of the bright contact zone to fill the internal holes. The morphological closure reconstruction operation uses the candidate regions of the bright contact zone as the marker image and the unfolded 2D image to be decomposed as the mask image. Morphological dilation is then performed on the candidate regions of the bright contact zone, using a circular structuring element with a radius of 3 pixels. After dilation, the result is compared with the unfolded 2D image to be decomposed pixel by pixel to obtain the reconstructed image. This dilation and minimum value operation is repeated until the reconstructed image no longer changes. After iterative convergence, a bright contact zone region that fills the internal holes and maintains boundary continuity is obtained. The above morphological closure reconstruction operation is performed on the candidate regions of the bright contact zone at each scale to obtain a bright contact zone region with enhanced continuity at the corresponding scale.

[0033] In specific implementation, please refer to Figure 3 Based on the morphological gradient distribution characteristics of the highlighted contact zone region at different scales, a morphological feature map of adjacent contact zones is constructed. For the highlighted contact zone region at multiple scales, the morphological gradient at each scale is calculated. The morphological gradient is the pixel difference between the dilation and erosion results of the highlighted contact zone region. For the highlighted contact zone region at each scale, a grayscale morphological dilation operation is performed using a circular structuring element with a radius of 3 pixels to obtain the dilation result; a grayscale morphological erosion operation is performed using a circular structuring element with a radius of 3 pixels to obtain the erosion result. Both the dilation and erosion results are images of the same size as the highlighted contact zone region. The grayscale value of each pixel in the dilation result is denoted as... The grayscale value of each pixel in the corrosion result is denoted as . ,in, Indicates the scale number. The value ranges from 1 to 15, corresponding to 15 scales for the radius of the circular structural element, ranging from 1 pixel to 15 pixels. Represents the row and column coordinates of a pixel within the highlight contact zone region. Morphological gradient value. Calculated using the following formula: Normalize all morphological gradient values ​​across all scales. Find the minimum value among the morphological gradient values ​​at all pixel locations across all scales. and maximum value For each scale and the position of each pixel Normalized gradient value The calculation method is as follows: After normalization, all gradient values ​​fall within the range. The normalized gradient values ​​are stacked sequentially in ascending order of scale to form a three-dimensional gradient distribution tensor. The three dimensions of the three-dimensional gradient distribution tensor are pixel row coordinates. Pixel column coordinates and scale The tensor is located in The element value at that position is During the stacking process, cubic spline interpolation is used to smooth the gradient values ​​along the scale direction. As interpolation nodes, the node values ​​are the normalized morphological gradient values ​​at the corresponding scale, for each fixed pixel position. A cubic spline interpolation function is constructed along the scale direction, with the natural boundary conditions of the interpolation nodes being of the second kind, i.e., the second derivative is zero at the endpoints. Four interpolation points are interpolated between adjacent original scales to obtain a gradient sequence with finer scale intervals, thereby improving the continuity of the 3D gradient distribution tensor along the scale direction. Along the increasing scale direction, the maximum gradient response value at each pixel location is extracted from the 3D gradient distribution tensor to obtain a maximum gradient response map. For each pixel location... Traversing the mesoscale dimension of the three-dimensional gradient distribution tensor Compare all values The size of the value is used to record the maximum gradient response value at that pixel location as the maximum value obtained during the traversal. All pixel locations A single-channel maximum gradient response map is constructed, with its size matching that of the original highlighted contact zone image. The maximum gradient response map and the two-dimensional unfolded image are then fused pixel-wise to generate a morphological feature map. The two-dimensional unfolded image is the unwound two-dimensional image after anisotropic diffusion filtering and adaptive histogram equalization, and after removing background interference from the restoration's edge contour. The maximum gradient response map and the two-dimensional unfolded image have the same number of rows and columns. Pixel-wise fusion uses a weighted average method, applying a weighted average to each pixel location. The pixel value of the morphological feature map at this location pass The calculation yielded that, For two-dimensional unfolded images in The grayscale values ​​at the specified locations are weighted by coefficients of 0.6 and 0.4, which are set according to the importance of the maximum gradient response map in characterizing the intensity of the contact zone edge. The morphological feature map simultaneously preserves the original grayscale information of the two-dimensional unfolded image and the gradient response intensity information of the highlighted contact zone region at multiple scales.

[0034] In specific implementation, please refer to Figure 4 In the morphological feature map, the continuous boundaries of adjacent contact zones are traced, and the length, width, and area quantization values ​​of the contact zones are calculated using topological analysis methods. A grayscale threshold is set in the morphological feature map, and the pixel values ​​of the morphological feature map are... The image is obtained by weighted averaging of the maximum gradient response map and the two-dimensional unfolded image. Represents pixel row coordinates, This represents the pixel column coordinates. The grayscale threshold is set to the 85th percentile of all pixel grayscale values ​​in the morphological feature map. It is calculated by sorting the grayscale values ​​of all pixels in the morphological feature map in ascending order and taking the grayscale value at the 85th percentile position as the grayscale threshold. Pixels with grayscale values ​​greater than the threshold are marked as contact zone seed points; pixels with grayscale values ​​less than or equal to the threshold are not marked.

[0035] Eight-neighbor connected component marking is performed on the contact zone seed points to obtain multiple initial contact zone connected components. The eight-neighbor connected component marking adopts a seed filling algorithm. Starting from an unvisited contact zone seed point, the eight neighboring pixels of the starting point are checked. Pixels that belong to the contact zone seed point among the neighboring pixels are added to the current connected component. The eight-neighbors of the newly added pixels are then checked again until the current connected component no longer expands. After marking one connected component, the next unvisited contact zone seed point is found as the new starting point, and the process of checking the eight-neighbors and expanding the connected component is repeated until all contact zone seed points are marked in their respective initial contact zone connected components, resulting in all initial contact zone connected components. Each initial contact zone connected component consists of a set of interconnected contact zone seed points.

[0036] Calculate the equivalent major axis and equivalent minor axis of the ellipse for each initial contact zone connected region. For each initial contact zone connected region, calculate the covariance matrix of all pixels within the connected region; the covariance matrix is... A matrix whose elements include the variance of the row coordinates of pixels within a connected region, the variance of the column coordinates of pixels, and the covariance between the row and column coordinates. The two eigenvalues ​​of the covariance matrix are calculated. and ,in, Length of the major axis of the equivalent ellipse pass The length of the minor axis of the equivalent ellipse is calculated. pass The calculations show that the length of the major axis of the equivalent ellipse is used as the initial length of the contact strip, and the length of the minor axis of the equivalent ellipse is used as the initial width of the contact strip.

[0037] Perform skeletonization extraction on each initial contact zone connected region to obtain a center skeleton line with a width of one pixel. Skeletonization employs the Zhang-Suen thinning algorithm, iteratively thinning the initial contact band connected components of the binary image. Each iteration consists of two sub-steps. The first sub-step marks and deletes boundary pixels that satisfy all of the following conditions: the number of foreground pixels in the eight neighbors of the pixel is between 2 and 6; the number of changes from background to foreground when traversing clockwise from a certain neighboring pixel of the pixel is 1; at least one of the pixels directly above, directly to the right, and directly below in the eight neighbors is background; and at least one of the pixels directly to the right, directly below, and directly to the left in the eight neighbors is background. The second sub-step marks and deletes boundary pixels that satisfy all of the following conditions: the number of foreground pixels in the eight neighbors of the pixel is between 2 and 6; the number of changes from background to foreground when traversing clockwise is 1; at least one of the pixels directly above, directly to the right, and directly to the left in the eight neighbors is background; and at least one of the pixels directly above, directly below, and directly to the left in the eight neighbors is background. Repeat the above two sub-steps until no pixels are deleted, and the remaining foreground pixels form a central skeleton line with a width of one pixel. Calculate the total pixel length of the central skeleton line as the precise length of the contact band. The total pixel length is the number of pixels that make up the central skeleton line. Each pixel contained in the central skeleton line is counted as one pixel unit in length. Add all pixel units together to obtain the precise length value.

[0038] The total number of pixels contained in each initial contact zone connected region is counted; this total number represents the number of all marked contact zone seed points within that region. The precise area of ​​the contact zone is calculated by multiplying the total number of pixels by the spatial resolution parameter of the 2D unfolded image. The spatial resolution parameter of the 2D unfolded image is the actual physical size corresponding to each pixel, which is read from the scan metadata of the 3D scan image and includes the physical spacing of pixels in the row direction. Physical spacing in the column direction Precise area pass The calculation yielded that, This represents the total number of pixels contained in the initial contact band connected region.

[0039] In practice, the three-dimensional surface coordinates of adjacent regions in the 3D scan image are obtained. The 3D scan image is acquired using an intraoral scanner while the restoration is in place, and includes the 3D coordinates of the triangular mesh and vertices of the restoration and adjacent teeth surfaces. The method for obtaining the 3D surface coordinates of adjacent regions is as follows: based on the pixel-to-3D vertex mapping relationship established during the generation of the 2D unfolded image, the 3D coordinate values ​​of all vertices corresponding to the adjacent regions of the restoration are indexed in the 3D scan image, and these vertex coordinate values ​​are combined into a set of 3D surface coordinates.

[0040] In the 3D scan image of the restoration, the 3D point cloud corresponding to the pixel positions in the 2D unfolded image is marked, obtaining a set of 3D point cloud coordinates. The mapping relationship between the 2D unfolded image and the 3D scan image is output by the conformal mapping algorithm used in the 2D unfolding process. This mapping relationship is stored in the form of a lookup table, recording the triangular facet index and centroid coordinates in the 3D scan image corresponding to the row and column positions of each pixel in the 2D unfolded image. For the position of each pixel on the central skeleton line, the nearest neighbor matching algorithm is used to retrieve the 2D image mapping record closest to the pixel position on the central skeleton line from the lookup table. The centroid coordinates of the triangular facet corresponding to this mapping record are used as the 3D point cloud coordinates of that pixel. The nearest neighbor matching algorithm uses Euclidean distance in the coordinate plane of the 2D unfolded image as the distance metric, selecting the mapping point with the smallest distance, thus obtaining the 3D point cloud coordinates corresponding to each pixel on the central skeleton line. The 3D point cloud coordinates corresponding to all pixels on the central skeleton line constitute the set of 3D point cloud coordinates.

[0041] Calculate the Euclidean distance between the 3D point cloud coordinates corresponding to each pixel on the central skeleton line, and sum all the Euclidean distances to obtain the 3D contact zone length. The central skeleton line consists of an ordered sequence of pixels, with connections between adjacent pixels. Let the total number of pixels on the central skeleton line be... , No. The 3D point cloud coordinates corresponding to each pixel are: , No. The 3D point cloud coordinates corresponding to each pixel are: , The value ranges from 1 to If the integer is true, then the Euclidean distance between the corresponding 3D points of two adjacent pixels is... The calculation formula is:

[0042]

[0043] Among them, the three-dimensional contact strip length It is obtained by summing the Euclidean distances between all adjacent points, i.e. .

[0044] Calculate the pixel width perpendicular to the central skeleton line within the initial contact zone's connected region. Based on the 3D point cloud coordinates corresponding to the pixel width, calculate the 3D distance along the normal direction. Take the average of all 3D distances along the normal direction as the 3D contact zone width. For each pixel on the central skeleton line, calculate the local tangent vector at that pixel. The local tangent vector is determined by the difference in 3D coordinates between the preceding and succeeding pixels on the central skeleton line. The direction perpendicular to the local tangent vector is the normal direction. Within the initial contact zone's connected region, starting from the pixel on the central skeleton line, detect pixel by pixel along the normal direction until reaching the boundary of the initial contact zone's connected region, recording the positions of the boundary pixels on both sides. Obtain the 3D point cloud coordinates corresponding to the boundary pixels on both sides, calculate the 3D Euclidean distance between the corresponding 3D point coordinates, and obtain the local value of the contact zone width at that skeleton point position. Traverse all pixels on the central skeleton line, calculate the local value of the contact zone width at each pixel position, and calculate the arithmetic mean of all local contact zone width values. The average value obtained is the 3D contact zone width. .

[0045] The area of ​​the surface patch enclosed by the 3D point cloud corresponding to all pixels within the initial contact zone's connected domain is calculated, and this surface patch area is taken as the 3D contact zone area. Each pixel within the initial contact zone's connected domain corresponds to a triangular facet region in the 3D scanned image. This triangular facet region is uniquely determined by the triangular facet index and centroid coordinates recorded in the mapping lookup table. For each pixel within the connected domain, the coordinates of the three vertices of its corresponding triangular facet are extracted from the 3D scanned image. The area of ​​the triangular facet is calculated using Heron's formula: the area of ​​a single triangular facet is equal to the square root of the product of half the sum of the three side lengths of the triangular facet and the product of the half perimeter and the individual side lengths. The 3D contact zone area is obtained by summing the areas of the triangular facets corresponding to all pixels within the initial contact zone's connected domain. Fourier transforms are performed on the three-dimensional contact strip length, width, and area to extract the low-frequency component amplitude values ​​in the frequency domain. The three-dimensional contact strip length... Treating the input sequence as a one-dimensional sequence of length 1, a one-dimensional discrete Fourier transform is applied. The frequency domain coefficients of the discrete Fourier transform are equal to the input sequence values ​​themselves. The amplitude of these frequency domain coefficients is extracted to obtain the frequency domain amplitude value of the contact strip length. The width of the three-dimensional contact strip Treating it as a one-dimensional sequence of length 1, a one-dimensional discrete Fourier transform is applied to extract the amplitude of the frequency domain coefficients, thus obtaining the frequency domain amplitude value of the contact band width. The three-dimensional contact area Treating it as a one-dimensional sequence of length 1, we apply a one-dimensional discrete Fourier transform to extract the amplitude of the frequency domain coefficients, thus obtaining the frequency domain amplitude value of the contact zone area. The three amplitude values ​​mentioned above correspond to the low-frequency components of their respective input quantities in the frequency domain, and the amplitude values ​​of the low-frequency components are the absolute values ​​of each input quantity.

[0046] The low-frequency component amplitude values ​​are weighted and summed according to preset weighting coefficients to generate a comprehensive adjacency tightness index. The preset weighting coefficients are set as follows: area weight 0.5, length weight 0.3, and width weight 0.2. The weighting coefficients are based on multi-factor statistical assessments of adjacency tightness in clinical practice: 200 clinical prosthodontic samples were collected, and three senior prosthodontists independently scored the contact area, length, and width of each sample, providing a comprehensive tightness level. Multiple linear regression analysis was used, with tightness level as the dependent variable and area, length, and width as independent variables. After normalization, the standardized regression coefficients showed a relative importance of 0.5 for area, 0.3 for length, and 0.2 for width. (Comprehensive Adjacency Tightness Index) The calculation method is as follows This weighted summation generates a single quantitative indicator for assessing the closeness of the contact relationship between the restoration and adjacent teeth.

[0047] In practice, consistency checks are performed on the quantized values ​​of the contact zone's length, width, and area. The standard deviation of the pixel grayscale values ​​within the initial contact zone's connected components is calculated. The initial contact zone's connected components are obtained from the morphological feature map through grayscale thresholding and eight-neighbor connected component labeling. The set of pixel locations covered by the initial contact zone's connected components in the morphological feature map is defined as follows: ,gather The number of pixels contained is For sets Each pixel position Obtain the pixel value at the given pixel location from the morphological feature map. ,in, Indicates the row number of the pixel. Indicates the column number of the pixel. Let the value be a real number. Calculate the set. The mean of all pixel values mean The calculation method is as follows The standard deviation of pixel grayscale values ​​within the initial contact band connected region. Calculated using the following formula:

[0048]

[0049] in, This represents the standard deviation of pixel grayscale values ​​within the initial contact band connected region. Represents a set The number of pixels in the middle. Representing morphological feature maps at pixel locations Pixel value at that location, Represents a set The mean of all pixel values. The standard deviation. The value was compared with a preset fluctuation threshold. The preset fluctuation threshold was set at 0.15, based on the following: Statistical analysis of the grayscale distribution of uniform and non-uniform contact zone samples in morphological feature maps showed that the mean standard deviation for 30 uniform contact zone samples was 0.08, with a maximum standard deviation not exceeding 0.12; in samples with localized abnormal contact, the standard deviation typically exceeded 0.18; therefore, 0.15 was selected as the critical value for distinguishing between uniform contact and localized abnormal contact. When the fluctuation value is less than the preset fluctuation threshold of 0.15, the quantized values ​​of the length, width, and area of ​​the contact zone are determined to be consistent. The process then proceeds directly to the three-dimensional spatial frequency domain mapping step, where the three-dimensional surface coordinates of adjacent regions in the three-dimensional scan image are obtained, followed by Fourier transform and weighted summation operations to generate a comprehensive adjacency density index. When the standard deviation... When the fluctuation threshold of 0.15 or higher is greater than or equal to the preset fluctuation threshold, it is determined that there are local anomalies in the quantized values ​​of the length, width, and area of ​​the contact strip. The initial connected component of the contact strip is then segmented, and the quantized values ​​of the length, width, and area are recalculated. Local minima are detected in the initial connected component of the contact strip. These local minima are used as seed points for segmentation, and a watershed segmentation algorithm is executed to divide the initial connected component of the contact strip into multiple sub-connected components. The detection method for local minima is as follows: within the pixel range covered by the initial connected component of the contact strip, a morphological H-minimum transform is performed on the pixel values ​​of the morphological feature map. The depth parameter in the morphological H-minimum transform... Set to 0.08, depth parameter The setting is based on the fact that the minimum contrast difference between the local gray-level minimum value and the surrounding pixels in the contact zone is usually between 0.05 and 0.10. A threshold of 0.08 is chosen as the threshold that can effectively filter out small noise fluctuations while retaining meaningful local minimum points. The specific operation of the morphological H-minimum transformation is to subtract the depth parameter from the pixel values ​​of the morphological feature map. The image after subtraction is morphologically reconstructed and compared with the original morphological feature map. Pixels that remain unchanged after reconstruction are marked as local minima. Detected local minima are used as seed points for segmentation, and each seed point is assigned a unique region label. The gradient magnitude of the morphological feature map is used as the terrain surface for the watershed segmentation algorithm. Region growing is performed starting from each seed point. During region growing, the terrain surface is submerged layer by layer in order of increasing gradient magnitude. When the grown regions of adjacent seed points meet, a watershed ridge is formed. Upon completion of segmentation, the initial contact zone is divided into multiple sub-connected regions, each corresponding to a local region where a seed point is located.

[0050] Calculate the length, width, and area quantization value for each sub-connected region. For the ... For each sub-connected region, calculate its precise length, precise width, and precise area, using the same method as for calculating the quantization value of the initial contact zone connected region: for the ... Skeletonization is performed on each sub-connected component to obtain its center skeleton line. The total pixel length of the center skeleton line of the sub-connected component is then calculated as the length quantization value of the sub-connected component. Calculate the length of the minor axis of the equivalent ellipse of the subconnected region as the quantized value of the width of the subconnected region. ; Calculate the area quantization value of the sub-connected region by counting the total number of pixels within the sub-connected region and combining it with the spatial resolution parameter. There are a total of Connected sub-domains The value range is 1 to Integer values. Calculate the mean and variance of multiple sets of subquantized values. Mean of length quantized values. pass The mean of the width quantization value was calculated. pass The mean of the area quantification value was calculated. pass The calculation is obtained. For each sub-connected region... Calculate the combined variance of the sub-connected regions. Comprehensive variance The calculation method is as follows Compare the combined variance of all connected sub-domains. The sub-connected region with the smallest overall variance is determined as the main contact zone region. The original contact zone length, width, and area quantization values ​​are replaced by the quantization values ​​of the main contact zone region. After replacement, the length, width, and area quantization values ​​of the main contact zone region are used as the precise length, precise width, and precise area of ​​the contact zone, replacing the contact zone length, width, and area quantization values ​​previously calculated directly from the initial contact zone connected region.

[0051] In some embodiments, after generating a quantitative index for assessing the closeness of adjacency relationships, the quantitative index is graded. A comprehensive closeness index of adjacency relationships is obtained from multiple standard dental prostheses. The standard dental prostheses are clinically collected 3D scan data after placement of the prostheses; the sample size is [number missing]. , No fewer than 100 cases. Clinical grading was performed on multiple standard dental prostheses to obtain clinical grading labels. Clinical grading was conducted by professional dentists with over five years of clinical experience in prosthodontics. The grading was based on the strength of the resistance encountered when floss passed between the prosthesis and adjacent teeth, and the translucency of the contact area as observed by the naked eye. The contact was categorized into three levels: too tight, normal, and too loose, corresponding to clinical grading label values ​​of 1, 2, and 3, respectively.

[0052] Using the adjacency density index as input features and clinical rating labels as output targets, a support vector machine (SVM) classifier is trained. The core architecture of the SVM classifier is as follows: the input layer receives a one-dimensional feature vector, which contains only the adjacency density index as a feature component; the kernel layer uses radial basis functions (RBFs) to calculate the similarity between the input samples and the support vectors. The RBF has the following form: ,in, Indicates the first The adjacency density index of each input sample. Indicates the first The adjacency density index of each support vector is calculated by considering the adjacency relationships of the support vectors. The radial basis function width parameter. The value is 0.5. The value is based on the fact that, in a one-dimensional feature space, the reciprocal median of the distance distribution between samples is approximately 0.5.

[0053] The sample spacing can be well mapped to the kernel function value space; the decision layer uses a one-to-many strategy to construct three binary classifiers, each trained on a clinical rating label, and the output is the classification score of the sample belonging to the corresponding level. The final output is the level with the highest classification score as the predicted level. The specific training steps of the support vector machine classifier are as follows: the adjacency relationship of standard dental prosthesis samples is combined with the clinical rating label to form a training sample set; the radial basis function width parameter is set. Set penalty parameters Penalty parameters The penalty for misclassification is set to the default standard value; the sequence minimum optimization algorithm is used to solve the support vector machine dual optimization problem, and the convergence accuracy is set to... After training, three classification decision functions and corresponding support vector sets are obtained. The comprehensive adjacency density index calculated for the prosthesis to be evaluated is input into the trained support vector machine classifier. The support vector machine classifier outputs three classification scores corresponding to the comprehensive adjacency density index. The three classification scores are compared, and the clinical rating label corresponding to the category with the highest score is the adjacency level of the prosthesis to be evaluated. This adjacency level is then output.

[0054] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A system for quantitatively evaluating the adjacency relationships of dental prostheses based on images, characterized in that, include: The image acquisition and preprocessing module acquires a three-dimensional scan image of the oral prosthesis in its in-place state, performs grayscale mapping and two-dimensional unfolding processing on the three-dimensional scan image, and obtains a two-dimensional unfolded image of the adjacent area of ​​the prosthesis. The image acquisition and preprocessing module performs multi-scale morphological decomposition on the two-dimensional unfolded image, extracts the structural elements of the adjacency relationship image at each scale, and separates the bright contact zone region through morphological opening and closing reconstruction operation. The feature map construction module constructs a morphological feature map of adjacent contact zones based on the morphological gradient distribution characteristics of the highlighted contact zone region at different scales. The topology analysis and calculation module traces the continuous boundaries of adjacent contact zones in the morphological feature map and uses topology analysis methods to calculate the length, width, and area quantification values ​​of the contact zones. The quantitative assessment mapping module maps the length, width, and area of ​​the contact strip to the three-dimensional spatial frequency domain of the restoration, generating a quantitative index for assessing the tightness of the adjacency relationship.

2. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 1, characterized in that, The specific steps of performing multi-scale morphological decomposition on the two-dimensional unfolded image, extracting the structural elements of the adjacency relationship image at each scale, and separating the bright contact zone region through morphological opening and closing reconstruction operations include: The two-dimensional unfolded image is subjected to grayscale morphological erosion operation. Circular structuring elements of different sizes are used to erode the image sequentially to obtain eroded images at multiple scales. A grayscale morphological dilation operation is performed on the eroded image at each scale, using a circular structuring element of the same size as the erosion operation for dilation, to obtain morphological operation results at multiple scales; Subtract the morphological unfolding operation results at each scale from the two-dimensional unfolded image to obtain candidate regions of bright contact zones at multiple scales; A morphological closure reconstruction operation is performed on the candidate region of the high-brightness contact zone to fill the voids inside the candidate region, thereby obtaining a high-brightness contact zone region with enhanced continuity.

3. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 2, characterized in that, Based on the morphological gradient distribution characteristics of the highlighted contact zone region at different scales, the specific steps for constructing the morphological feature map of adjacent contact zones include: Calculate the morphological gradient of the bright contact zone region at each scale, where the morphological gradient is the pixel difference between the dilation result and the erosion result of the bright contact zone region; The morphological gradients at all scales are normalized, and the normalized gradient values ​​are stacked in ascending order of scale to form a three-dimensional gradient distribution tensor. Along the direction of increasing scale, the maximum gradient response value at each pixel location is extracted from the three-dimensional gradient distribution tensor to obtain the maximum gradient response map. The maximum gradient response map is fused with the two-dimensional unfolded image at the pixel level to generate the morphological feature map.

4. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 3, characterized in that, When the normalized gradient values ​​are stacked in ascending order of scale, cubic spline interpolation is used to smooth the gradient values ​​in the scale direction.

5. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 3, characterized in that, The specific steps for tracing the continuous boundaries of adjacent contact zones in the morphological feature map and calculating the length, width, and area quantification values ​​of the contact zones using topological analysis methods include: In the morphological feature map, a grayscale threshold is set, and pixels with a grayscale value greater than the grayscale threshold are marked as contact zone seed points. The contact zone seed points are then marked with eight neighbor connected components to obtain multiple initial contact zone connected components. Calculate the equivalent major axis length and equivalent minor axis length of the ellipse for each initial contact zone connected region. Use the equivalent major axis length as the initial length of the contact zone and the equivalent minor axis length as the initial width of the contact zone. A skeletonization extraction operation is performed on each initial contact band connected region to obtain a central skeleton line with a single pixel width, and the total pixel length of the central skeleton line is calculated as the precise length of the contact band. The total number of pixels contained in each initial contact zone's connected region is counted, and the precise area of ​​the contact zone is calculated by combining this with the spatial resolution parameters of the two-dimensional unfolded image.

6. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 5, characterized in that, The specific steps for mapping the quantified values ​​of the length, width, and area of ​​the contact band to the three-dimensional spatial frequency domain of the restoration to generate a quantitative index for assessing the tightness of the adjacency relationship include: Obtain the three-dimensional surface coordinates of adjacent regions in the three-dimensional scan image, and project the precise length, precise width, and precise area of ​​the contact strip onto the three-dimensional surface coordinates to obtain the three-dimensional contact strip length, three-dimensional contact strip width, and three-dimensional contact strip area; Fourier transform is performed on the length, width, and area of ​​the three-dimensional contact strip to extract the amplitude values ​​of the low-frequency components in the frequency domain; The amplitude values ​​of the low-frequency components are weighted and summed according to preset weighting coefficients to generate a comprehensive adjacency density index.

7. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 6, characterized in that, The preset weighting coefficients have an area weight of 0.5, a length weight of 0.3, and a width weight of 0.

2.

8. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 6, characterized in that, The specific steps for obtaining the three-dimensional surface coordinates of adjacent regions in the three-dimensional scan image of the restoration, and projecting the precise length, precise width, and precise area of ​​the contact band onto the three-dimensional surface coordinates to obtain the three-dimensional contact band length, three-dimensional contact band width, and three-dimensional contact band area include: In the three-dimensional scan image of the restoration, mark the three-dimensional point cloud corresponding to the pixel position of the two-dimensional unfolded image to obtain a set of three-dimensional point cloud coordinates; Calculate the Euclidean distance between the three-dimensional point cloud coordinates corresponding to each pixel on the central skeleton line, and sum all the Euclidean distances to obtain the length of the three-dimensional contact strip; Calculate the pixel width in the direction perpendicular to the central skeleton line within the connected domain of the initial contact strip, calculate the three-dimensional distance along the normal direction based on the three-dimensional point cloud coordinates corresponding to the pixel width, and take the average value of all three-dimensional distances in the normal direction as the three-dimensional contact strip width. The area of ​​the surface patch enclosed by the three-dimensional point cloud corresponding to all pixels within the initial contact zone is calculated, and the area of ​​the surface patch is taken as the area of ​​the three-dimensional contact zone.

9. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 8, characterized in that, The 3D point cloud coordinates corresponding to each pixel on the central skeleton line are obtained through a nearest neighbor matching algorithm.

10. The image-based quantitative evaluation system for the adjacency relationship of dental prostheses according to claim 1, characterized in that, After obtaining the two-dimensional unfolded image of the adjacent region of the restoration, and before performing multi-scale morphological decomposition on the two-dimensional unfolded image, a preprocessing operation is also included. The specific steps of the preprocessing operation include: Anisotropic diffusion filtering is applied to the two-dimensional unfolded image to remove image noise and retain edge information, resulting in a filtered image. Adaptive histogram equalization is performed on the filtered image to enhance the contrast between adjacent contact zones and background regions, resulting in a contrast-enhanced image. The edge contour of the restoration is identified in the contrast-enhanced image, and the pixel grayscale value of the outer region of the edge contour of the restoration is set to zero to obtain a two-dimensional unfolded image to be decomposed after removing background interference.

Images