A digital modeling and virtual reconstruction system for cultural relics

Abstract

The application discloses a cultural relic digital modeling and virtual reconstruction system, and belongs to the technical system of image processing; curvature shock spectrum decomposition is performed on a three-dimensional point cloud of a cultural relic; a point cloud curvature field is regarded as a non-steady-state signal; a multi-scale curvature shock detection operator is constructed; micro trace features and scanning noise of the cultural relic are separated; and a hierarchical feature anchor point set and geometric parameters thereof are output; the output hierarchical feature anchor point set is used as a seed; topological constraint crystallization grid growth reconstruction is performed; micro trace feature line anchors are anchored as grain boundary constraints; global curved surface grid construction is completed on the premise of maintaining sharp features; and a feature faithful reconstruction grid is output; cross-modal light field-geometry joint verification is performed on the output feature faithful reconstruction grid; through residual analysis of structured light response and grid normal rendering response, a micro trace distortion area is located, and adaptive grid correction is driven; and a final cultural relic digital model is output.

Description

Technical Field

[0001] This invention relates to image processing technology systems, specifically to a system for digital modeling and virtual reconstruction of cultural relics. Background Technology

[0002] When implementing existing digital modeling and virtual reconstruction schemes for cultural relics, the geometric depth of core identification features such as oracle bone inscriptions, bronze inscriptions, and jade ridges is often only tens of micrometers. In point clouds, these features appear as sharp local curvature abrupt changes. However, general completion algorithms, such as those based on Poisson reconstruction or smoothing interpolation, treat these micro-traces as noise or smooth them into a blurry slope, erasing key information that can be used for dating and interpretation. This results in the defect of smoothing out sharp features and microscopic traces. Summary of the Invention

[0003] The purpose of this invention is to provide a digital modeling and virtual reconstruction system for cultural relics to solve the technical problems mentioned in the background art.

[0004] The objective of this invention can be achieved through the following technical solutions: A system for digital modeling and virtual reconstruction of cultural relics includes: The 3D point cloud processing module for cultural relics performs curvature shock spectral decomposition on the 3D point cloud of cultural relics, treats the curvature field of the point cloud as an unsteady signal, constructs a multi-scale curvature shock detection operator, separates the micro-trace features of cultural relics from scanning noise, and outputs a hierarchical feature anchor set and its geometric parameters. Feature-fidelity reconstruction mesh building module: Using the output hierarchical feature anchor point set as seed crystals, it performs crystallization mesh growth and reconstruction with topological constraints, anchors micro-trace feature lines as grain boundary constraints, completes global surface mesh construction while maintaining sharp features, and outputs feature-fidelity reconstruction mesh. The verification and correction module for the digital model of cultural relics performs cross-modal light field-geometry joint verification on the output feature-fidelity reconstructed mesh. Through residual analysis of structured lighting response and mesh normal rendering response, it locates micro-trace distortion areas and drives adaptive mesh correction to output the final digital model of cultural relics.

[0005] Furthermore, the original three-dimensional point cloud set of the artifact surface is obtained, and for the i-th sampling point in the point cloud set... Construct its local neighborhood point set; calculate the number of points in the local neighborhood point set, and obtain the j-th neighboring point in the local neighborhood point set. Extract the unit normal vector of the sampling point. and neighboring points The unit normal vector; based on the extracted unit normal vector, calculate the sampling points. The normal deviation tensor at that point.

[0006] Furthermore, eigenvalue decomposition is performed on the normal deviation tensor to obtain the first, second, and third eigenvalues; using the first and third eigenvalues, the sampling points are calculated. The intensity of the curvature shock wave at that location.

[0007] Furthermore, when constructing the multi-scale curvature shock wave detection operator, the total number of scale levels r is set, and an index k for the scale levels is introduced, and Set the basic neighborhood radius Calculate the neighborhood radius at the k-th scale level. ; At each scale level k, with the corresponding neighborhood radius The local neighborhood is reconstructed, and the curvature shock wave intensity is repeatedly calculated. The repeatedly calculated curvature shock wave intensities are sorted and combined in chronological order to obtain a multi-scale shock wave intensity sequence. With neighborhood radius Construct sampling points with the independent variable. The scale-shock response curve is calculated, and the length of the continuous scale interval in the scale-shock response curve where the curvature shock wave intensity is greater than the preset shock wave threshold is defined as the sampling point. The duration and width of the shock wave.

[0008] Furthermore, when performing feature classification and noise removal based on shock wave duration and shock wave intensity, based on sampling points... The shock wave duration width and the curvature shock wave intensity at the maximum scale are used to perform classification and determination. For sampling points determined to be valid micro-trace features Along the direction of the third eigenvector, the maximum indentation depth within the local neighborhood point set is measured and defined as the sampling point. The corresponding geometric depth parameters; integrate all effective micro-trace feature points and output a hierarchical feature anchor point set.

[0009] Furthermore, when constructing the grain boundary constraint plane for the notch-type feature, when the feature type label is determined to be a notch-type linear groove feature, adjacent spatially collinear anchor points are extracted from the hierarchical feature anchor point set, and the adjacent anchor points are connected to form the m-th feature line chain; the unit tangential vector of the feature line chain is calculated; the cross product operation is performed using the first feature vector and the unit tangential vector to calculate the normal vector of the grain boundary constraint plane; the grain boundary constraint plane is constructed with the feature line chain as the center line and the normal vector as the normal.

[0010] Furthermore, when constructing a double-sided constraint wedge for ridge-type features, when the feature type label is determined to be a ridge-type protruding feature, adjacent spatially collinear anchor points are extracted from the hierarchical feature anchor point set, and the adjacent anchor points are connected to form the nth ridge chain. Obtain the Euclidean distance between adjacent anchor points in the ridge chain, denoted as the adjacent anchor point spacing; and obtain the output geometric depth parameters, and calculate the wedge angle of the double-sided constrained wedge based on the geometric depth parameters and the adjacent anchor point spacing. Using the ridge chain as the edge and the wedge angle as the opening angle, a double-sided constrained wedge is constructed along the symmetry plane of the first eigenvector.

[0011] Furthermore, when constructing the spherical constraint region for point-like imprint features, when the feature type label determines the point-like imprint feature, the sphere is constructed with the corresponding feature anchor point's three-dimensional spatial coordinates as the center and twice the geometric depth parameter as the radius. A spherical constrained region.

[0012] Furthermore, during the crystallization grid growth and constraint truncation, each seed crystal starts from the three-dimensional spatial coordinates of the feature anchor point and expands outward layer by layer along the initial growth direction to form triangular facets. During the expansion process, the growth front of the triangular facets is detected in real time to see if it touches the boundary of the grain boundary constraint plane, the double-sided constraint wedge, or the spherical constraint region. When the growth front touches any of the above constraint boundaries, the truncation and reflection rules are executed.

[0013] Furthermore, when calculating the true structured illumination response, the first... A real image illuminated by a structured light source; for the u-th pixel in the real image, its gray value is extracted as the real illumination intensity; By iterating through all structured light sources, the actual illumination intensity of the u-th pixel under different light sources is combined to construct the actual illumination response vector of the u-th pixel. When calculating the cross-modal residual and locating the micro-trace distortion region, calculate the cross-modal residual between the real illumination response vector of the u-th pixel and the virtual rendering response vector of the corresponding u-th mesh vertex; When the cross-modal residual is greater than the preset residual judgment threshold, the grid region corresponding to the u-th pixel is determined to have geometric distortion and is marked as a micro-trace distortion region.

[0014] Compared to existing solutions, the beneficial effects achieved by this invention are: This invention successfully preserves core identification features with geometric depths of only tens of micrometers, such as oracle bone inscriptions, bronze inscriptions, and jade ridges, through multi-scale curvature shock spectral decomposition, preventing the loss of crucial dating and interpretation information. Utilizing the persistence of shock wave response at multiple scales and the single-scale attenuation characteristics of scanning noise in real micro-traces, high-frequency scanning noise is adaptively eliminated without manual intervention, significantly improving the signal-to-noise ratio of feature extraction. The output hierarchical feature anchor set not only contains precise three-dimensional spatial coordinates but also carries feature type labels, shock wave intensity, and geometric depth parameters, providing rigorous grain boundary anchoring data for subsequent topologically constrained crystallization mesh growth, ensuring the geometric fidelity of sharp features during subsequent mesh reconstruction.

[0015] This invention introduces grain boundary constraint planes, double-sided constraint wedges, and spherical constraint regions to rigidly anchor the geometric boundaries of micro-mark features such as oracle bone inscriptions, bronze inscriptions, and jade ridges. This forces the mesh to be truncated at the feature edges while maintaining tangential continuity, completely overcoming the fatal flaw of traditional smooth interpolation algorithms that treat micro-marks as noise and smooth them out as blurry slopes. The angle of the double-sided constraint wedges is precisely calculated by geometric depth and anchor point spacing, and the spherical constraint region is directly driven by geometric depth. This ensures that the reconstructed mesh is highly consistent with the actual artifact in key identification features such as mark depth, ridge sharpness, and pit diameter, providing reliable geometric data support for subsequent dating and interpretation. While strictly protecting micro-mark features, isotropic Delaunay triangulation is used to fill free regions, ensuring the smoothness of the overall surface of the artifact and the quality of the mesh. This avoids the problem of global mesh distortion caused by local feature protection, providing high-quality mesh input for subsequent light field-geometric joint verification.

[0016] This invention introduces a realistic structured lighting response as a physical standard, amplifying geometric errors into significant optical residuals, thus achieving ultra-sensitive localization of micro-trace distortion areas. By minimizing the lighting residual energy function to drive mesh correction, the normal distribution of the virtual mesh is forced to approximate the physical reflection characteristics of real cultural relics. This ensures that, in the final model's virtual display, the shadow depth of oracle bone inscriptions, the edge highlights of bronze inscriptions, and the gloss transition of jade ridges are all completely consistent with the visual effect of real cultural relics under the same lighting conditions. During the adaptive correction process, a mesh smoothing regularization term is introduced, effectively avoiding mesh tearing, normal flipping, or surface unevenness caused by pursuing ultimate optical matching. This ensures that the final digital model of the cultural relic, while possessing extremely high micro-trace fidelity, still maintains a high-quality, watertight, and smooth global surface topology. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1This is a flowchart illustrating the operation of a digital modeling and virtual reconstruction system for cultural relics according to the present invention. Detailed Implementation

[0019] 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.

[0020] like Figure 1 As shown, this invention is a digital modeling and virtual reconstruction system for cultural relics, comprising: The 3D point cloud processing module for cultural relics performs curvature shock wave spectral decomposition on the 3D point cloud of cultural relics, treating the point cloud curvature field as an unsteady signal, constructing a multi-scale curvature shock wave detection operator, separating the micro-trace features of the cultural relics from scanning noise, and outputting a hierarchical feature anchor point set and its geometric parameters; specifically: When acquiring the 3D point cloud of an artifact and calculating the local neighborhood normal deviation tensor, a high-precision 3D scanner is used to obtain the original 3D point cloud set P on the surface of the artifact. For the i-th sampling point in the point cloud set P... Let i be the index of the sampling point, and construct its local neighborhood point set. ; Calculate the local neighborhood point set Number of points in And obtain the local neighborhood point set. The j-th neighborhood point j is the index of the neighboring points; extract sampling points unit normal vector and neighboring points unit normal vector Based on the above vectors, calculate the sampling points. Normal deviation tensor at point The calculation formula is: The normal deviation tensor is a real symmetric matrix, specifically... With respect to the normal deviation tensor Eigenvalue decomposition yields three eigenvalues, which are denoted as the first eigenvalue. Second eigenvalue and the third eigenvalue And satisfy Eigenvalue decomposition is performed using existing conventional techniques; the specific implementation steps are not detailed here. The first eigenvalue is then extracted. The corresponding first feature vector and the third eigenvalue The corresponding third eigenvector ; in, In the formula, As the first candidate vector, , The first adjoint matrix The Column vector, First adjoint matrix For matrix The transpose of the cofactor matrix; It is a 3-order identity matrix; Indicates the first column vector The Euclidean norm; The first candidate vector The Euclidean norm; For operators that take extreme values; In the formula, As the third candidate vector, , The third adjoint matrix The Column vector, ; Indicates the first column vector The Euclidean norm; The third candidate vector The Euclidean norm; It should be noted that by converting the changes in the normals of the local neighborhood of the point cloud into tensor form, we can accurately capture the drastic changes in the normal direction of the micro-trace region, providing a tensor algebra basis for distinguishing between real micro-traces and smooth noise.

[0021] When calculating the intensity of a single-scale curvature shock wave based on the normal deviation tensor, the first eigenvalue is used. and the third eigenvalue Calculate sampling points Curvature shock wave intensity at The calculation formula is: In the formula, To prevent the regularization constant from having a denominator of zero, a value of is taken. ; It should be noted that the intensity of curvature shock waves The calculation formula is essentially a normalized extremum anisotropy index. Its design consists of three parts: a numerator (difference measure), a denominator (normalization and suppression mechanism), and a regularization constant. The specific principle is explained as follows: Numerator : Capture the maximum anisotropy of normal changes; In the normal deviation tensor In the eigenvalue system, the first eigenvalue The third eigenvalue represents the maximum variance of the change in the normal vector within the local neighborhood, i.e., the direction of the most drastic change in the normal vector. The minimum variance characterizing the change in the normal vector is the direction in which the change in the normal is most gradual. For linear groove features like inscriptions, such as those on oracle bones: the normal to the cross-section perpendicular to the direction of the inscription is drastically deflected. The value is extremely large, while the normal remains almost unchanged in the direction parallel to the direction of the groove. Minimal, corresponding to the difference It is extremely large and can keenly capture ridge-valley mutations with linear characteristics; For isotropic noise or spherical / corner features: the perturbation or variation amplitude of the normal in all directions tends to be consistent, i.e. At this time, the difference Approaching zero; Therefore, by extracting the range between the maximum and minimum eigenvalues, the numerator term algebraically quantifies the direction selectivity of the local curvature field, which is the core criterion for distinguishing linear micro-traces from isotropic noise. denominator Achieve relative intensity normalization and flat region suppression; If only the absolute difference is used This approach faces two serious engineering flaws: first, the absolute magnitudes of feature values ​​vary greatly depending on the material and scanning precision of the artifacts, making it impossible to set a globally uniform judgment threshold; second, in absolutely flat areas, i.e. Even tiny floating-point calculation errors can cause the difference to oscillate wildly; introducing a denominator term plays the following key role: Normalization of high-response regions: when the sampling point is located in the actual micro-trace region, Significantly greater than zero and much greater than The denominator is approximately The formula degenerates into ;at this time Strictly mapped to Within the dimensionless range, the shock threshold has universality and physical interpretability across cultural relics and scanning equipment; Automatic suppression of low-response regions: When the sampling point is located in an absolutely flat region or a pure noise region, It approaches zero; at this point, the denominator changes from a regularization constant. Dominance, making It tends towards 0; this mechanism is mathematically eliminated. The risk of numerical overflow caused by indeterminate expressions is mitigated, and the shock wave intensity in flat areas is automatically suppressed to zero to prevent the calculation rounding error from being misjudged as a micro-trace feature.

[0022] Intensity of curvature shock waves With the preset shock threshold When a comparison is performed, At that time, the sampling points were initially determined. Located on the curvature shock wave surface, it serves as a candidate point for micro-trace features. The preset shock wave threshold can be determined by pre-scanning a standard micro-trace calibration block with known true values ​​to obtain the distribution of the normal deviation tensor eigenvalues ​​of the calibration sample. The median of the lower quartile of the true micro-trace region's eigenvalues ​​and the upper quartile of the non-micro-trace region's eigenvalues ​​can be used as the threshold. Alternatively, it can be determined based on conventional techniques in the field, such as adaptive determination based on global statistical distribution. The determination of other thresholds involved in this embodiment is the same as the determination of the shock wave threshold, and will not be elaborated further. It needs to be explained that the intensity of curvature shock waves This reflects the maximum difference in the degree of curvature of a local surface in various directions. The edges of oracle bone inscriptions or bronze inscriptions will produce extremely large anisotropic curvature, thus exhibiting high shock wave intensity, while the normal perturbation caused by scanning noise is usually isotropic and has a lower shock wave intensity.

[0023] When constructing a multi-scale curvature shock wave detection operator, the total number of scale levels r is set, and an index k for the scale levels is introduced. Set the basic neighborhood radius Calculate the neighborhood radius at the k-th scale level. The calculation formula is: ; At each scale level k, with the corresponding neighborhood radius The local neighborhood is reconstructed, and the curvature shock wave intensity is repeatedly calculated. The repeatedly calculated curvature shock wave intensities are sorted and combined in chronological order to obtain a multi-scale shock wave intensity sequence. With neighborhood radius Construct sampling points with the independent variable. Scale-shock response curve Calculation scale-shock response curve In this context, the length of a continuous scale interval where the intensity of the curvature shock wave exceeds a preset shock wave threshold is defined as a sampling point. Shock wave duration width ; It should be noted that through multi-scale analysis, real artifact micro-traces, such as scratches tens of micrometers deep, will maintain a stable high shock wave response across multiple continuous scales, while high-frequency scanning noise only exhibits shock wave characteristics at extremely small scales (microscopic neighborhoods) and decays rapidly as the scale increases, thus achieving physical separation of micro-traces and noise in the scale domain.

[0024] When performing feature classification and noise removal based on shock wave duration and intensity, based on sampling points Shock wave duration width and curvature shock wave intensity at the largest scale Perform classification and determination: when and At that time, determine the sampling point It is a linear groove feature of the scoring type; when and At that time, determine the sampling point It is a ridge-like protrusion feature; when and At that time, determine the sampling point It is characterized by dot-like imprints; when and At that time, determine the sampling point To scan for and remove noise; For sampling points determined to be valid micro-trace features Along the third eigenvector The direction of measurement of the local neighborhood point set The maximum depth of the indentation within is defined as the sampling point. Corresponding geometric depth parameters ; Integrate all valid micro-trace feature points to output a hierarchical feature anchor point set A, where each element of the set is represented as... ,in, Sampling points Corresponding feature type label; It should be noted that by jointly determining the duration and intensity of the shock wave, tiny scratches that were mistakenly identified as noise by the general algorithm were completely eliminated, while the geometric depth parameters of the features were preserved, providing precise seed locations and depth constraint boundaries for grain boundary constraints and crystallization mesh growth in subsequent steps.

[0025] It should be noted that by treating the curvature field of the point cloud as an unsteady signal and introducing the shock wave detection theory from fluid mechanics, the microscopic sharp features of cultural relics can be accurately extracted, classified, and quantified from the mixed noise of the three-dimensional point cloud.

[0026] In this embodiment of the invention, multi-scale curvature shock spectral decomposition successfully preserves core identification features with geometric depths of only tens of micrometers, such as oracle bone inscriptions, bronze inscriptions, and jade ridges, preventing the loss of key dating and interpretation information. This overcomes the limitations of traditional Poisson reconstruction or smooth interpolation algorithms, which treat minute curvature abrupt changes as noise. By utilizing the persistence of shock wave response at multiple scales of real micro-traces and the single-scale attenuation characteristics of scanning noise, high-frequency scanning noise is adaptively eliminated without manual intervention, significantly improving the signal-to-noise ratio of feature extraction. The output hierarchical feature anchor set not only contains precise three-dimensional spatial coordinates but also carries feature type labels, shock wave intensity, and geometric depth parameters, providing strict grain boundary anchoring data for subsequent topologically constrained crystallization mesh growth, ensuring the geometric fidelity of sharp features during subsequent mesh reconstruction.

[0027] Feature-fidelity reconstruction mesh construction module: Using the output hierarchical feature anchor point set as seed crystals, it performs topological constraint-based crystallization mesh growth and reconstruction, anchors micro-trace feature lines as grain boundary constraints, and completes global surface mesh construction while maintaining sharp features, outputting a feature-fidelity reconstruction mesh; specifically: When initializing the seed crystal and determining the growth direction, the output hierarchical feature anchor point set is obtained, and the first step is extracted. Three-dimensional spatial coordinates of each feature anchor point And the corresponding feature type label; the three-dimensional spatial coordinates of each feature anchor point. Initialize as a seed crystal; The index of the feature anchor points in the hierarchical feature anchor point set; Obtain the calculated normal deviation tensor The first eigenvector corresponding to the first eigenvalue , the first feature vector Set the initial growth direction of the seed crystal; obtain the set basic neighborhood radius. ; It should be noted that by directly initializing the identified micro-trace feature points as seed crystals and guiding their growth along their principal curvature direction (first feature vector), the starting point and direction of the mesh reconstruction are ensured to be strictly anchored on the real physical micro-trace, thus avoiding feature shift caused by the traditional algorithm starting growth from uniform sampling points.

[0028] When constructing the grain boundary constraint plane for a notch-type feature, if the feature type label indicates a notch-type linear groove feature, adjacent spatially collinear anchor points are extracted from the hierarchical feature anchor point set, and these adjacent anchor points are connected to form the m-th feature line chain. , m is the index of the characteristic line chain; calculate the characteristic line chain. unit tangential vector Using the first eigenvector With unit tangential vector Perform cross product operation to calculate grain boundary constraint plane. normal vector The calculation formula is: ; with characteristic line chain Using the center line and the normal vector Construct a grain boundary constraint plane with the normal as the normal. ; Where, when the first feature vector With unit tangential vector When parallel (i.e., the cross product results in a zero vector), a degenerate case is handled, specifically: Calculate the grain boundary constraint plane normal vector First, parallelism detection is performed; then the first feature vector is calculated. With unit tangential vector cross product modulus Set a parallelism threshold. In this embodiment, we take ; When the cross product modulus is long When determining the first feature vector With unit tangential vector If they are not parallel, calculate the grain boundary constraint plane using the original formula. normal vector ; When the cross product modulus is long When determining the first feature vector With unit tangential vector If the cross product is parallel and the result is a zero vector or an approximately zero vector, the normal vector of the constraint plane cannot be uniquely determined. In this case, the process of handling the degenerate case begins. It should be noted that the first feature vector Characteristic of normal deviation tensor The direction corresponding to the largest eigenvalue, i.e., the principal direction of the most drastic change in the normal, is the unit tangential vector. Characteristic line chain The direction of the scratch; if the two are parallel, it means that the direction of the scratch is exactly the same as the direction of the maximum change of the normal. Although this situation rarely occurs in actual artifact micro-marks, it may occur at the intersection of oblique scratches, intersecting scratches, or in areas of high curvature of curved surfaces. It must be properly handled, specifically including: First-level regression: Introducing a second feature vector Substitute; When detected and When parallel, extract the normal deviation tensor. The second eigenvalue The corresponding second feature vector Due to the normal deviation tensor Since it is a real symmetric matrix, its eigenvectors are mutually orthogonal; therefore, the second eigenvector is... It must be related to the first eigenvector Orthogonal; Calculate the second eigenvector With unit tangential vector cross product modulus ; when At that time, with the second eigenvector Replace the first feature vector The grain boundary constraint plane is calculated using the following alternative formula. normal vector : ; It should be noted that the second eigenvalue The corresponding second feature vector The secondary principal direction characterizing the change of normal; in scuff-type features, when When following the direction of the engraving, It usually points in the transverse cross-sectional direction of the notch, that is, the width direction, therefore... and The normal vector obtained by the cross product It can still effectively define the lateral constraint plane of the groove, and the physical meaning is reasonable.

[0029] Second-level backtracking: Introducing a third feature vector Substitute; When the second feature vector With unit tangential vector cross product modulus At that time, that is also with Parallel, at this time Simultaneously parallel to and This is theoretically impossible in three-dimensional space because and Orthogonal; however, considering floating-point precision issues, this level of backoff still needs to be set to extract the normal deviation tensor. The third eigenvalue The corresponding third eigenvector ; Due to the three eigenvectors , , To form a three-dimensional orthogonal basis, when Simultaneously parallel to and hour, Inevitably and Orthogonal, therefore It must be a non-zero vector; calculate the normal vector using the following alternative formula. : ; Third-level rollback: Introducing local surface normal vectors structure; When numerical anomalies prevent the aforementioned three levels of feature vectors from producing effective cross products—a case of extreme degradation—feature line chains are introduced. Local surface normal vector at the center point Local surface normal vector By using feature anchor points The 3D point cloud within the neighborhood is obtained by least-squares plane fitting. Calculate the local surface normal vector With unit tangential vector The cross product is used to calculate the normal vector using the following alternative formula. : ; It should be noted that the local surface normal vector It is independent of the normal deviation tensor. The geometric quantities of the eigenvector system are determined by the spatial distribution of the local point cloud; due to the feature line chains of the indentation class. Extending along the surface, unit tangential vector Located within the tangent plane, while the local surface normal vector Perpendicular to the tangent plane, the two are naturally orthogonal, therefore the cross product is... It must be a non-zero vector, which fundamentally guarantees the unique determinability of the constrained plane normal vector.

[0030] It needs to be explained that, physically, the notch-like feature is a narrow groove, the edge of which is the area most easily smoothed during mesh reconstruction. By constructing a grain boundary constraint plane that is perpendicular to the notch direction and orthogonal to the principal curvature direction, it is equivalent to setting up physical baffles on both sides of the groove, forcing the subsequent mesh to stop crossing when it grows to this point, thereby locking the width and depth boundaries of the notch.

[0031] When constructing a bi-faceted constrained wedge for ridge-type features, if the feature type label determines it to be a ridge-type protruding feature, spatially collinear adjacent anchor points are extracted from the hierarchical feature anchor point set, and these adjacent anchor points are connected to form the nth ridge chain. , where n is the index of the ridge chain; Obtain the ridge chain The Euclidean distance between adjacent anchor points is denoted as the anchor point spacing *l*; and the output geometric depth parameters are obtained. Based on geometric depth parameters Calculate the double-sided constrained wedge based on the distance l between adjacent anchor points. wedge angle The calculation formula is: ; With ridge chain The ridge line is the angle subtended by the wedge. The opening angle is along the first eigenvector. Construct a double-sided constrained wedge based on the plane of symmetry. ; It should be noted that ridge-like features, such as the ridges of jade artifacts, have sharp tops, which traditional algorithms would smooth into rounded arcs. The double-sided constraint wedge precisely calculates the angle by using geometric depth and anchor point spacing, forming symmetrical inclined constraints on both sides of the ridge, forcing the mesh to form sharp angles at the ridge, perfectly reproducing the sharp protrusion of the artifact.

[0032] When constructing a spherical constraint region for point-like imprint features, when the feature type label determines the point-like imprint feature, the sphere is constructed with the corresponding feature anchor point's three-dimensional spatial coordinates as the center and twice the geometric depth parameter as the radius. A spherical constrained region ; It needs to be explained that dot-like marks, such as tiny casting pores on the surface of bronze artifacts, or dot-like corrosion pits, have abrupt changes in depth in all directions; the spherical constraint region forms a closed three-dimensional protective shield, ensuring that the depth and diameter of the dot-like micro-marks are not stretched or deformed by the surrounding freely growing mesh during the mesh growth process.

[0033] During crystallization mesh growth and constraint truncation, each seed crystal originates from the three-dimensional spatial coordinates of its feature anchor point. Initially, triangular facets are expanded outward layer by layer along the initial growth direction; during the expansion process, the growth front of the triangular facets is monitored in real time to determine whether they touch the grain boundary constraint plane. Double-sided constraint wedge or spherical constrained region The boundary; When the growth front touches any of the above constraint boundaries, the truncation and reflection rules are executed. Specifically, the growth direction is reflected or truncated by the constraint boundary, and the normal of the newly generated triangular facet maintains tangential continuity with the constraint geometry at the constraint boundary. It should be noted that this "crystallization" growth mechanism draws on the grain boundary growth theory in materials science, treating micro-trace constraints as insurmountable "grain boundaries." The mesh patches, like crystals, naturally stop and adhere at the constraint boundaries, fundamentally eliminating the feature erasure phenomenon caused by mesh patches crossing micro-trace boundaries in the general Poisson reconstruction algorithm.

[0034] When performing isotropic filling of free regions and iteration termination, for planes not constrained by grain boundaries... Double-sided constraint wedge and spherical constrained region The covered free-growing region is filled with a mesh using an isotropic Delaunay triangulation algorithm, which is a conventional existing technical solution. The specific implementation steps will not be elaborated here. Set the maximum side length limit of the filled triangle to be The growth process continues to iterate until the growth fronts of all the crystal seeds intersect each other, or the growth fronts reach the bounding box boundary of the three-dimensional point cloud of the cultural relic, at which point growth stops and the feature-fidelity reconstruction mesh M is output. It should be noted that, under the premise of ensuring that the micro-trace features are strictly constrained and protected, Delaunay triangulation is used to efficiently fill the flat or gently sloping free regions. This not only ensures the integrity and smoothness of the global surface, but also avoids the generation of poor-quality meshes that are too narrow or distorted by limiting the maximum side length. The final feature-fidelity reconstructed mesh will serve as the geometric benchmark for subsequent cross-modal optical field-geometry joint verification.

[0035] It is worth noting that by using the extracted micro-trace features as seed crystals to perform topologically constrained crystallization mesh growth and reconstruction, the sharp micro-trace features are physically-level geometrically locked and reconstructed with high fidelity during the 3D mesh reconstruction process.

[0036] In this embodiment of the invention, by introducing grain boundary constraint planes, double-sided constraint wedges, and spherical constraint regions, the geometric boundaries of micro-mark features such as oracle bone inscriptions, bronze inscriptions, and jade ridges are rigidly anchored. This forces the mesh to be truncated at the feature edges while maintaining tangential continuity, completely overcoming the fatal flaw of traditional smooth interpolation algorithms that treat micro-marks as noise and smooth them out as blurry slopes. The angle of the double-sided constraint wedges is precisely calculated by geometric depth and anchor point spacing, and the spherical constraint region is directly driven by geometric depth. This ensures that the reconstructed mesh is highly consistent with the actual artifact in key identification features such as inscription depth, ridge sharpness, and pit diameter, providing reliable geometric data support for subsequent dating and interpretation. While strictly protecting micro-mark features, isotropic Delaunay triangulation is used to fill free regions, ensuring the smoothness of the overall surface of the artifact and the quality of the mesh. This avoids the problem of global mesh distortion caused by local feature protection, providing high-quality mesh input for subsequent light field-geometric joint verification.

[0037] The artifact digitization model verification and correction module performs cross-modal light field-geometry joint verification on the output feature-fidelity reconstructed mesh. Through residual analysis of the structured lighting response and mesh normal rendering response, it locates micro-distortion areas and drives adaptive mesh correction, outputting the final artifact digitization model. Specifically: During the acquisition of multimodal reference data and initialization, the output feature-fidelity reconstructed mesh is obtained, and the vertex set of the feature-fidelity reconstructed mesh is extracted, denoted as mesh vertex set V. The x-th mesh vertex in mesh vertex set V is then obtained. initial unit normal vector x represents the index of a grid vertex; simultaneously, a sequence of images of the artifact surface acquired using multi-directional structured light sources in a real physical environment is obtained, denoted as the real light field image set. ; Set the total number of structured light sources to L, and obtain the first... The unit vector of the incident light direction of a structured light source. ,in, ; It should be noted that by simultaneously acquiring geometric mesh data and real physical light field data, a virtual geometry-real physics benchmark was established for subsequent cross-modal joint verification, ensuring the physical authenticity of the verification process.

[0038] When calculating the true structured illumination response, from the real light field image set Extract the first A real image illuminated by a structured light source. For real images The u-th pixel in the image, where u is the pixel index, has its grayscale value extracted as the true illumination intensity. ; Iterate through all L structured light sources, combine the actual illumination intensities of the u-th pixel under different light sources, and construct the actual illumination response vector of the u-th pixel. True illumination response vector The Each component is, that is, ; It should be explained that the real light response vector accurately records the optical scattering, self-shadowing, and multiple reflection characteristics of micro-traces on the surface of cultural relics under real physical conditions, and is a physical standard for evaluating the geometric fidelity of micro-traces.

[0039] When calculating the mesh normal rendering response, in the virtual rendering environment, lighting conditions are constructed that are completely consistent with the real physical environment, that is, using the same L incident light direction unit vectors. The feature-fidelity reconstructed mesh is projected onto the same image plane as the real image. For the x-th mesh vertex corresponding to the u-th pixel after projection... Based on the Lambertian diffuse lighting model, the x-th mesh vertex is calculated at the... Virtual rendering lighting intensity under structured light sources The calculation formula is: In the formula, The `()` function is used to remove negative lighting from backlighting sources; it iterates through all L structured light sources and constructs the virtual rendering response vector for the x-th mesh vertex. Virtual rendering response vector The Each component is... ; It should be noted that virtual rendering response depends only on the geometric normals of the mesh. If the geometry of the mesh is distorted in the micro-trace area, such as in the depth or normal direction, the rendered lighting response will deviate significantly from the real physical lighting response.

[0040] When calculating cross-modal residuals and locating micro-trace distortion regions, the true illumination response vector of the u-th pixel is calculated. , and the virtual rendering response vector of the corresponding u-th mesh vertex Cross-modal residuals The calculation formula is: ; Set residual judgment threshold and cross-modal residuals With residual judgment threshold When a comparison is performed, At that time, if the grid region corresponding to the u-th pixel is determined to have geometric distortion, it is marked as a micro-distortion region. ; It needs to be explained that cross-modal residuals quantify the difference between real physical lighting and virtual geometry rendering. Micro-trace distortion areas are small defect areas where the transition of local normals is unnatural or the depth is not fully met due to constraint truncation during crystallization mesh growth.

[0041] When driving adaptive mesh correction, an illumination residual energy function is constructed for the mesh vertices within the micro-distortion region. The calculation formula is: In the formula, Let x be the unit normal vector to be determined after correction at the x-th mesh vertex. For the x-th grid vertex Adjacent grid vertices, For the x-th grid vertex The set of adjacent grid vertices, This represents the grid smoothing regularization weight coefficient, with a value range of (0,1). Represents grid vertices With adjacent grid vertices The Euclidean distance between them; Gradient descent method is used to analyze the illumination residual energy function. The solution is minimized, and the position coordinates and normal vectors of the mesh vertices within the micro-distortion region are iteratively updated until the illumination residual energy function is reached. The value is less than the preset energy convergence threshold, and the grid correction is completed; the gradient descent method implemented is an existing conventional technical solution, and the specific implementation steps are not described here; It should be noted that the first term of the energy function corresponds to the data fidelity term, which forces the correction of the mesh normal rendering response to approximate real physical lighting. The second term corresponds to the regularization term, which prevents the mesh from becoming excessively distorted or having holes during the correction process, thereby achieving the optimal balance between optical fidelity and geometric topological integrity.

[0042] When performing iterative verification and outputting the final model, the corrected mesh is used as a feature-fidelity reconstruction mesh M. The steps of calculating the mesh normal rendering response, calculating the cross-modal residual and locating the micro-distortion region, and driving adaptive mesh correction are repeated until the cross-modal residual of all pixels is calculated. If all values ​​are less than or equal to the residual judgment threshold, the iteration terminates and the final digital model of the cultural relic is output.

[0043] It should be noted that through closed-loop iterative verification, the slight geometric deviations that may remain after a single correction are completely eliminated, ensuring that the final output digital model can present the same micro-trace optical texture as the real cultural relic under any lighting conditions.

[0044] It should be explained that by performing cross-modal optical field-geometry joint verification on the output mesh, the virtual geometric mesh is subjected to optical-level precision verification and adaptive closed-loop correction using real physical optical field data.

[0045] Traditional methods rely solely on the geometric distance from the point cloud to the mesh to assess reconstruction quality, failing to detect minute deviations in the normal direction of micro-traces. This invention, by introducing a realistic structured lighting response as a physical standard, amplifies geometric errors into significant optical residuals, achieving ultra-sensitive localization of micro-trace distortion areas. By minimizing the lighting residual energy function to drive mesh correction, the normal distribution of the virtual mesh is forced to approximate the physical reflection characteristics of real artifacts. This ensures that, in virtual display, the shadow depth of oracle bone inscriptions, the edge highlights of bronze inscriptions, and the luster transition of jade ridges are completely consistent with the visual effect of real artifacts under the same lighting. During adaptive correction, a mesh smoothing regularization term is introduced, effectively avoiding mesh tearing, normal flipping, or surface unevenness caused by pursuing ultimate optical matching. This ensures that the final digital artifact model, while possessing extremely high micro-trace fidelity, still maintains a high-quality, watertight, and smooth global surface topology.

[0046] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0047] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0048] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module described above can be implemented in hardware or in the form of hardware plus software functional modules.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system for digital modeling and virtual reconstruction of cultural relics, characterized in that, include: The 3D point cloud processing module for cultural relics performs curvature shock spectral decomposition on the 3D point cloud of cultural relics, treats the curvature field of the point cloud as an unsteady signal, constructs a multi-scale curvature shock detection operator, separates the micro-trace features of cultural relics from scanning noise, and outputs a hierarchical feature anchor set and its geometric parameters. Feature-fidelity reconstruction mesh building module: Using the output hierarchical feature anchor point set as seed crystals, it performs crystallization mesh growth and reconstruction with topological constraints, anchors micro-trace feature lines as grain boundary constraints, completes global surface mesh construction while maintaining sharp features, and outputs feature-fidelity reconstruction mesh. The verification and correction module for the digital model of cultural relics performs cross-modal light field-geometry joint verification on the output feature-fidelity reconstructed mesh. Through residual analysis of structured lighting response and mesh normal rendering response, it locates micro-trace distortion areas and drives adaptive mesh correction to output the final digital model of cultural relics.

2. The system for digital modeling and virtual reconstruction of cultural relics according to claim 1, characterized in that, Obtain the original 3D point cloud set of the artifact's surface. For the i-th sampling point in the point cloud set... Construct its local neighborhood point set; calculate the number of points in the local neighborhood point set, and obtain the j-th neighboring point in the local neighborhood point set. Extract the unit normal vector of the sampling point. and neighboring points The unit normal vector; based on the extracted unit normal vector, calculate the sampling points. The normal deviation tensor at that point.

3. The system for digital modeling and virtual reconstruction of cultural relics according to claim 2, characterized in that, Eigenvalue decomposition is performed on the normal deviation tensor to obtain the first, second, and third eigenvalues; the sampling points are then calculated using the first and third eigenvalues. The intensity of the curvature shock wave at that location.

4. The system for digital modeling and virtual reconstruction of cultural relics according to claim 3, characterized in that, When constructing a multi-scale curvature shock wave detection operator, the total number of scale levels r is set, and an index k for the scale levels is introduced. ; Set the basic neighborhood radius Calculate the neighborhood radius at the k-th scale level. ; At each scale level k, with the corresponding neighborhood radius The local neighborhood is reconstructed, and the curvature shock wave intensity is repeatedly calculated. The repeatedly calculated curvature shock wave intensities are sorted and combined in chronological order to obtain a multi-scale shock wave intensity sequence. With neighborhood radius Construct sampling points with the independent variable. The scale-shock response curve is calculated, and the length of the continuous scale interval in the scale-shock response curve where the curvature shock wave intensity is greater than the preset shock wave threshold is defined as the sampling point. The duration and width of the shock wave.

5. The system for digital modeling and virtual reconstruction of cultural relics according to claim 4, characterized in that, When performing feature classification and noise removal based on shock wave duration and intensity, based on sampling points The shock wave duration width and the curvature shock wave intensity at the maximum scale are used to perform classification and determination. For sampling points determined to be valid micro-trace features Along the direction of the third eigenvector, the maximum indentation depth within the local neighborhood point set is measured and defined as the sampling point. The corresponding geometric depth parameters; integrate all effective micro-trace feature points and output a hierarchical feature anchor point set.

6. The system for digital modeling and virtual reconstruction of cultural relics according to claim 5, characterized in that, When constructing the grain boundary constraint plane of the notch-type feature, when the feature type label is determined to be a notch-type linear groove feature, the adjacent spatially collinear anchor points are extracted from the hierarchical feature anchor point set, and the adjacent anchor points are connected to form the m-th feature line chain. Calculate the unit tangential vector of the characteristic line chain; calculate the normal vector of the grain boundary constraint plane by performing a cross product operation between the first eigenvector and the unit tangential vector; construct the grain boundary constraint plane with the characteristic line chain as the center line and the normal vector as the normal.

7. The system for digital modeling and virtual reconstruction of cultural relics according to claim 6, characterized in that, When constructing a double-sided constraint wedge with ridge-type features, if the feature type label is determined to be a ridge-type protruding feature, adjacent spatially collinear anchor points are extracted from the hierarchical feature anchor point set, and the adjacent anchor points are connected to form the nth ridge chain. Obtain the Euclidean distance between adjacent anchor points in the ridge chain, denoted as the adjacent anchor point spacing; and obtain the output geometric depth parameters, and calculate the wedge angle of the double-sided constrained wedge based on the geometric depth parameters and the adjacent anchor point spacing. Using the ridge chain as the edge and the wedge angle as the opening angle, a double-sided constrained wedge is constructed along the symmetry plane of the first eigenvector.

8. The system for digital modeling and virtual reconstruction of cultural relics according to claim 7, characterized in that, When constructing a spherical constraint region for point-like imprint features, when the feature type label determines the point-like imprint feature, the sphere is constructed with the corresponding feature anchor point's three-dimensional spatial coordinates as the center and twice the geometric depth parameter as the radius. A spherical constrained region.

9. A digital modeling and virtual reconstruction system for cultural relics according to claim 8, characterized in that, When performing crystallization mesh growth and constraint truncation, each seed crystal starts from the three-dimensional spatial coordinates of the feature anchor point and expands outward layer by layer along the initial growth direction to form triangular facets. During the expansion process, the growth front of the triangular facets is detected in real time to see if it touches the boundary of the grain boundary constraint plane, the double-sided constraint wedge, or the spherical constraint region. When the growth front touches any of the above constraint boundaries, the truncation and reflection rules are executed.

10. A digital modeling and virtual reconstruction system for cultural relics according to claim 9, characterized in that, When calculating the true structured illumination response, the first image is extracted from the set of true light field images. A real image illuminated by a structured light source; for the u-th pixel in the real image, its gray value is extracted as the real illumination intensity; By iterating through all structured light sources, the actual illumination intensity of the u-th pixel under different light sources is combined to construct the actual illumination response vector of the u-th pixel. When calculating the cross-modal residual and locating the micro-trace distortion region, calculate the cross-modal residual between the real illumination response vector of the u-th pixel and the virtual rendering response vector of the corresponding u-th mesh vertex; When the cross-modal residual is greater than the preset residual judgment threshold, the grid region corresponding to the u-th pixel is determined to have geometric distortion and is marked as a micro-trace distortion region.

Images