Cultural heritage three-dimensional digital reconstruction display method and system

Abstract

The invention discloses a cultural heritage three-dimensional digital reconstruction display method and system, and relates to the technical field of computer vision, and the method comprises the steps: constructing a multi-dimensional radiation distribution data set containing an observation angle and radiation intensity; calculating a surface displacement rate and an observation angle conversion rate of the radiation peak value, and determining a highlight migration characteristic according to a ratio of the two; generating a mask and decoupling to generate an initial diffuse reflection texture and material parameter map; further performing orthogonal decomposition on the difference vector based on a reference illumination direction, and performing anisotropic compensation on the texture by using a vertical component to generate a target texture; rendering the initial three-dimensional geometric model according to the target texture and the material parameter map to generate a three-dimensional digital reconstruction model; through physical optical decoupling and vector compensation, the problems of dynamic highlight misjudgment and inherent texture loss are effectively solved, and high-fidelity three-dimensional digital reconstruction is realized.

Description

Methods and Systems for 3D Digital Reconstruction and Display of Cultural Heritage Technical Field

[0001] This invention relates to the field of computer vision technology, specifically to a method and system for the three-dimensional digital reconstruction and display of cultural heritage. Background Technology

[0002] In the field of digital preservation of cultural heritage, three-dimensional reconstruction using multi-view image sequences has become a core means of permanently preserving the historical information of cultural relics. This technology usually involves first constructing a geometric model of the cultural relic through photogrammetry or laser scanning, and then mapping the color information of the images taken from multiple angles onto the surface of the model to generate a digital archive with realistic visual effects.

[0003] However, unlike rough stone or earthen artifacts, artifacts such as glazed porcelain, gilded bronze statues, or lacquered wood have highly reflective surfaces. When the acquisition equipment moves around the artifact to take pictures, the specular highlights generated by the ambient light source are not fixed at specific coordinates on the surface of the object. Instead, they optically drift on the surface of the artifact as the viewing angle changes continuously. Because current technology lacks the ability to deeply perceive and decouple this dynamic lighting characteristic, it often mechanically misjudges these dynamic bright spots that slide with the viewing angle as the static inherent color of the object's surface.

[0004] During the texture blending stage, moving highlights are mistakenly imprinted on the final texture map, forming false white patches or visual ghosting, which seriously damages the real texture of the cultural relic. If a brightness threshold is simply introduced to cut in order to remove the light spots, it is very easy to mistakenly identify the inherent light-colored texture of the cultural relic (such as the white background of blue and white porcelain or the gold and silver wire inlay pattern) as highlight noise and erase it as well, causing irreversible loss of the key semantic information of the cultural relic.

[0005] Therefore, how to accurately distinguish between optical interference that moves with the viewing angle and inherent texture that is fixed in position by analyzing the inherent logical patterns of the data, so as to completely preserve the real material details while eliminating false light spots, is the core technical problem that urgently needs to be solved in the field of high-fidelity 3D reconstruction of cultural relics. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for three-dimensional digital reconstruction and display of cultural heritage.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] In a first aspect, this invention discloses a method for three-dimensional digital reconstruction and display of cultural heritage, comprising the following steps:

[0009] Obtain the initial 3D geometric model of the target object to be reconstructed and the corresponding multi-view image sequence;

[0010] Pixel data from multi-view image sequences are mapped to surface units of an initial 3D geometric model to construct a multidimensional radiation distribution dataset containing observation angle and radiation intensity dimensions.

[0011] Based on a multidimensional radiation distribution dataset, the surface displacement rate of the radiation intensity peak of the surface unit and the corresponding observation angle transformation rate are calculated.

[0012] Based on the ratio of surface displacement rate to observation angle change rate, the spectrophotometric migration characteristic data of the surface unit are determined.

[0013] A high-brightness confidence mask is generated based on the high-brightness migration feature data, and a weighted decoupling process is performed on the multidimensional radiation distribution dataset based on the high-brightness confidence mask to generate an initial diffuse texture map and a material parameter map.

[0014] In the color space, a reference lighting direction vector is determined, and the difference vector between the initial diffuse texture map and the multi-view image sequence in the color space is constructed.

[0015] The difference vector is orthogonally decomposed based on the reference lighting direction vector, and the anisotropic compensation of the initial diffuse texture map is performed based on the vertical component obtained from the decomposition to generate the target texture map.

[0016] The initial 3D geometric model is rendered based on the target texture map and material parameter map to generate a 3D digital reconstruction model and display it.

[0017] Secondly, this invention discloses a three-dimensional digital reconstruction and display system for cultural heritage, comprising:

[0018] The data acquisition module is used to acquire the initial three-dimensional geometric model of the target object to be reconstructed and the corresponding multi-view image sequence;

[0019] The radiation distribution construction module is used to map pixel data in multi-view image sequences to surface units of an initial three-dimensional geometric model, and construct a multi-dimensional radiation distribution dataset containing the observation angle dimension and the radiation intensity dimension.

[0020] The rate calculation module is used to calculate the surface displacement rate of the radiation intensity peak of the surface unit and the corresponding observation angle transformation rate based on the multidimensional radiation distribution dataset.

[0021] The specular feature determination module is used to determine the specular migration feature data of surface units based on the ratio of surface displacement rate to observation angle change rate.

[0022] The texture decoupling module is used to generate a specular confidence mask based on specular migration feature data, and to perform weighted decoupling processing on the multidimensional radiation distribution dataset based on the specular confidence mask to generate an initial diffuse texture map and a material parameter map.

[0023] The lighting analysis module is used to determine the reference lighting direction vector in the color space and construct the difference vector between the initial diffuse texture map and the multi-view image sequence in the color space.

[0024] The texture compensation module is used to perform orthogonal decomposition on the difference vector based on the reference lighting direction vector, and to perform anisotropic compensation on the initial diffuse texture map according to the vertical component obtained by decomposition to generate the target texture map.

[0025] The rendering and display module is used to render the initial 3D geometric model based on the target texture map and material parameter map, generate a 3D digital reconstruction model, and display it visually.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] 1. A specular migration feature operator based on the ratio of surface displacement rate to observation angle change rate was constructed. Utilizing the physical law that inherent textures are anchored to the geometric surface (ratio approaches 0) while specular highlights slide with the viewing angle (ratio is significant), an optical kinematic filter independent of color space was constructed. This enables the system to accurately capture specular areas from a dynamic behavior perspective through appearances with similar brightness, completely eliminating ghosting and visual double images left on the surface of the 3D model due to the movement of light spots. It is particularly suitable for the reconstruction of highly reflective cultural relics such as glazed porcelain and gilded statues.

[0028] 2. The system introduces the difference vector orthogonal decomposition technique in the color space. By decomposing the texture residual into parallel components (representing the gain of light intensity) and vertical components (representing the deviation of chromaticity information), the system can accurately remove the brightness interference that causes the highlights. At the same time, it uses an anisotropic compensation strategy to backfill the vertical components into the texture. This mechanism ensures that while removing specular reflections, the hue and saturation information of the misjudged areas are completely recovered, thus ensuring the integrity of the fine semantic information on the surface of the cultural relics (such as the gold and silver wire inlay texture).

[0029] 3. A self-calibration mechanism for illumination vectors based on principal component analysis (PCA) is proposed. This mechanism can use the parallel components of the highlights that were removed in the initial calculation as statistical samples to deduce the true principal axis direction of the light source and correct the reference illumination vector accordingly. This closed-loop feedback design using residual correction parameters enables the system to automatically converge to the optimal solution without the need for optical calibration equipment. This effectively prevents texture color distortion or residual light spots caused by illumination estimation errors and reduces the hardware dependence on the acquisition environment. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 is an overall block diagram of the method of Embodiment 1 of the present invention;

[0032] Figure 2 is a flowchart of the overall execution of the method in Embodiment 1 of the present invention;

[0033] Figure 3 is a schematic diagram of the principle of specular migration feature detection in the method of Embodiment 1 of the present invention;

[0034] Figure 4 is an overall block diagram of the system of Embodiment 2 of the present invention. Detailed Implementation

[0035] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0036] Application Overview: In the field of high-fidelity digital reconstruction of cultural heritage, especially for artifacts with complex optical surface characteristics such as glazed porcelain and gilded statues, the construction of relightable three-dimensional digital assets is regarded as the core standard for achieving permanent preservation and interactive display. The generation of such high-fidelity digital twins is essentially a process of decoupling the observed radiation signal into the inherent diffuse albedo and the environmental specular reflection component at the optical physics level. That is, through the constraints of multi-viewpoint imaging geometry, using the conservation of photometric properties and the Fresnel effect, the microscopic scattering properties of the material surface are accurately restored, thereby reproducing the real texture of the artifact under different lighting conditions in digital space.

[0037] However, existing photogrammetric techniques are generally based on the assumption of ideal diffuse reflection, lacking a verification mechanism for the consistency of optical response between visual observation signals and physical surface properties. This leads to the inability to accurately identify optical drift and false texture confusion caused by light illuminating curved artifacts. Optical drift manifests as specular highlights having extremely high radiation intensity, but their spatial position shifts significantly with slight changes in the viewing angle, lacking geometric anchoring characteristics. False texture confusion manifests as the inherent light-colored patterns of the artifact (such as white glaze and gilding) having RGB values ​​extremely similar to highlights, causing the algorithm to misjudge inherent textures as noise signals during highlight removal. Consequently, a strict physical causal relationship cannot be established between pixel brightness changes and viewpoint movement, resulting in false detection of highlight areas or accidental deletion of texture details, thus affecting the color reproduction and semantic integrity of the 3D reconstruction model.

[0038] For example, when acquiring data from multiple viewpoints on a piece of blue-and-white underglaze red porcelain, traditional 3D reconstruction systems can only handle low-frequency diffuse lighting through conventional pixel-weighted averaging, but cannot distinguish whether high-frequency highlight signals are accompanied by dynamic shifts in viewing angle. Furthermore, when encountering white glaze areas with colors similar to the highlights, the system only cuts them based on statistical thresholds of brightness, failing to detect the relative static state of the radiation peak relative to the surface geometry under multiple viewpoints. Specifically, the system misjudges false highlights that slide with the viewing angle as static stains and bakes them onto the texture map to form ghosting, or misclassifies static bright white glaze as specular reflection and performs incorrect anisotropic repair, resulting in ghosting spots or permanent loss of fine textures in the reconstructed model, making it impossible to form a material representation that conforms to physical optics.

[0039] If the above problems are not resolved, the 3D digitization system will continue to lose its ability to resolve complex optical surfaces. In particular, the failure to decouple optical drift will cause ambient light noise to mix into texture maps, making digital cultural relics unable to adapt to the dynamic lighting of virtual exhibition halls, thereby weakening the realism of their interactive display. At the same time, the failure to correct texture confusion will cause irreversible erasure of the historical information of cultural relics, making the 3D model unable to bear the microscopic details required for cultural relic identification, and ultimately causing the digital archives to lose their academic value as a source of information. Therefore, the inaccuracy of optical resolution will systematically hinder the construction of a high-fidelity digital cultural relic database and affect the quality of cultural heritage digitization protection projects.

[0040] Example 1:

[0041] As shown in Figures 1-3, the method for three-dimensional digital reconstruction and display of cultural heritage includes the following steps:

[0042] Step S1: Obtain the initial 3D geometric model of the target object to be reconstructed and the corresponding multi-view image sequence;

[0043] In practice, this step mainly consists of three logical sub-processes: digital acquisition of geometric structures, full coverage capture of photometric information, and adaptive quality assessment based on frequency domain entropy.

[0044] Specifically, to construct a high-precision initial 3D geometric model, this embodiment preferably employs a handheld structured light 3D scanner or a ground-based LiDAR to perform a 3D all-around scan of the cultural heritage target (such as glazed porcelain or gilded Buddha statues). The scanning device acquires high-density 3D point cloud data of the object's surface by projecting coded gratings or laser pulses onto the surface. Subsequently, the system uses the Poisson Surface Reconstruction algorithm to transform the discrete point cloud data into a continuous triangular mesh model. In this process, the system not only records the vertex geometric coordinates of each surface unit (triangular facet) but also simultaneously calculates and stores the normal vector of each surface unit, providing a geometric reference for subsequent illumination angle calculations.

[0045] Meanwhile, when acquiring multi-viewpoint image sequences, this embodiment uses a high-resolution digital camera to perform multi-angle and multi-height surround shooting around the target object, ensuring that the overlap rate of the image sequence is higher than 60% to meet the requirements of subsequent feature matching. After acquisition, the system uses the Structure from Motion (SfM) algorithm to solve the image sequence, accurately obtaining the camera intrinsic parameters (focal length, principal point) and extrinsic parameters (rotation matrix, translation vector) corresponding to each frame, thereby establishing the spatial mapping relationship between two-dimensional image pixels and three-dimensional geometric models.

[0046] It is worth noting that, in order to prevent blurry images caused by hand tremors, focus failure, or rapid movement during the acquisition process from interfering with subsequent highlight analysis, this embodiment pre-sets image quality perception logic based on frequency domain features before the data enters the processing pipeline. This logic is not based on simple pixel contrast, but on the energy distribution characteristics of the image in the frequency domain space.

[0047] In actual operation, the system performs a Fast Fourier Transform (FFT) on each acquired multi-view image frame, transforming it from the spatial domain to the frequency domain. In the frequency domain, sharp images have higher and more widely distributed high-frequency components, while the high-frequency energy of blurry images is significantly attenuated. The system quantifies the image sharpness level by calculating the information entropy of the frequency domain amplitude spectrum. Assuming the normalized probability distribution of the frequency domain amplitude spectrum is... Then the information entropy of the image The calculation is as follows:

[0048]

[0049] Based on the calculated information entropy index The system further generates the distribution construction weights for the frame image using a pre-defined sigmoid mapping function. For example, when a frame image experiences information entropy due to motion blur... Below the preset quality threshold (e.g.) When this happens, the system will automatically reduce the weights of the distribution of that frame image to [a lower value]. They are even marked as invalid; while for keyframes with extremely high clarity, their weights are set to... .

[0050] This ensures that only high-quality visual information can be included in the subsequent multidimensional radiation distribution dataset, thereby blocking texture reconstruction noise caused by low-quality data at the source and achieving a robust closed loop from physical acquisition to data preparation.

[0051] Step S2: Map the pixel data in the multi-view image sequence to the surface units of the initial three-dimensional geometric model to construct a multi-dimensional radiation distribution dataset containing the observation angle dimension and the radiation intensity dimension;

[0052] In this embodiment, step S2 is a bridge connecting two-dimensional visual information and three-dimensional geometric structure. Its core lies in constructing a high-dimensional data container that can completely preserve the dynamic changes of illumination.

[0053] Specifically, the system first traverses each surface unit (typically represented as a triangular facet in practical engineering) in the initial 3D geometric model and calculates its geometric center coordinates and normal vector. To address the self-occlusion problem on complex artifact surfaces (such as deep carvings and openwork structures), the system does not simply map images from all camera views to the surface unit. Instead, it introduces a visibility filtering mechanism based on depth buffering (Z-buffer) or ray casting. Under this mechanism, the system emits a virtual ray from the optical center of each camera towards the geometric center of the surface unit. If the ray intersects with other facets of the model along its propagation path, the camera view is considered occluded and removed from the valid camera set. This process ensures that the subsequently acquired photometric data truly originates from the surface unit, rather than from occlusion objects.

[0054] After determining the effective camera set, this embodiment utilizes a pinhole camera model to perform inverse projection. The system reads the camera pose parameters (including rotation matrix) corresponding to each effective image frame. With translation vector and intrinsic parameter matrix Using the projection formula The three-dimensional geometric center coordinates of the surface unit Mapped to the two-dimensional imaging plane of each camera, precisely indexed to the corresponding pixel coordinates. And extract the pixel radiation intensity value (i.e., RGB color value or brightness value) at that location.

[0055] To eliminate the quantization error caused by the discrete pixel grid, this embodiment does not use simple nearest neighbor interpolation, but instead performs bilinear interpolation or bicubic interpolation. Specifically, the system reads the projected coordinates. Surrounding or The system calculates sub-pixel level radiance values ​​using distance weights within the pixel neighborhood. Furthermore, if the projected area covers multiple pixels in the image (i.e., the texture sampling rate is lower than the screen pixel rate), the system automatically triggers multi-level mipmap sampling to prevent aliasing artifacts such as moiré patterns from contaminating the radiance data.

[0056] Subsequently, the system enters the geometric feature calculation stage. To quantify the variation of illumination with viewing angle, the system constructs an observation line-of-sight vector connecting the optical centers of each effective camera with the geometric centers of the surface units. Based on this vector, the system further calculates its dot product with the surface unit normal vector, derives the cosine of the angle between them, and converts it into an observation angle value. This physical quantity precisely describes the angle at which the camera observes the geometric state of the point.

[0057] Ultimately, instead of compressing the data into a single color value, the system constructs a high-dimensional, multi-dimensional radiation distribution dataset. For each surface unit, the system pairs its observed angle values ​​across all effective viewpoints with pixel radiation intensity values, forming a series of binary sequences (e.g., ...). These sequences are not randomly stacked, but rather indexed and stored according to observation angle or time series. This data structure is essentially a discretized slice of the bidirectional reflectance distribution function (BRDF), which fully records the illumination response characteristics of the surface point under different viewpoints, providing a complete data foundation for subsequent steps.

[0058] Step S3: Based on the multidimensional radiation distribution dataset, calculate the surface displacement rate of the radiation intensity peak of the surface unit and the corresponding observation angle transformation rate;

[0059] In this embodiment, step S3 quantifies the motion sensitivity of the light spot relative to the viewpoint by analyzing the derivative characteristics of the photometric signal in the time and spatial dimensions. To ensure the accuracy of the physical meaning of the calculation, this step does not employ a simple image plane optical flow method, but instead performs rigorous geodesic analysis based on three-dimensional manifold geometry.

[0060] Specifically, to capture the movement trend of the light spot on the object's surface, the system first needs to define a topological neighborhood for each surface unit to be analyzed on the initial 3D geometric model. This neighborhood construction is not based on Euclidean distance, but rather on the topological connectivity of the mesh. The system uses a half-edge data structure to quickly retrieve adjacent faces that share vertices with the current surface unit, and then uses a K-ring expansion algorithm (e.g., setting K=3 or K=5, depending on the model's resolution) to determine a local curved surface region. This topological neighborhood not only defines the spatial boundary of the light spot search but also ensures that subsequent displacement calculations strictly conform to the geometric undulations of the artifact's surface.

[0061] After determining the search range, the system delves into the multidimensional radiation distribution dataset, executing peak path construction logic along the time axis of the observation times. The system sequentially reads adjacent observation times (denoted as...). and The data slices are used to retrieve the surface locations with the highest radiation intensity values ​​within the aforementioned topological neighborhood, and these locations are marked as peak points. With peak point .

[0062] Considering that sensor thermal noise may cause single-point extreme value drift, this embodiment introduces a local energy centroid correction algorithm when determining the peak point. Specifically, the system does not directly define the single sampling point with the highest radiation intensity as... Instead, it takes that maximum value as the center and selects a tiny [point / area]. The topological neighborhood is used to calculate the weighted geometric center of the radiation intensity of all points within the neighborhood, which is then used as the corrected sub-voxel peak position. This process significantly improves the smoothness of spot path tracing and avoids velocity calculation divergence caused by isolated noise points.

[0063] At this point, the system performs a data saturation integrity check: if detected... and When the pixel radiation intensity values ​​at a given location all reach the sensor's maximum range (e.g., 255), i.e., they are in a flat-top saturation state, the system will identify the risk of derivative calculation failure and automatically trigger an anomaly flag, skipping subsequent rate calculations to avoid division by zero or numerical explosion errors.

[0064] Assuming the data is valid, the system then calculates the physical rate of change in both dimensions in parallel. In the surface space dimension, the system does not directly calculate... and The straight-line distance (Euclidean distance) between the two points is used because a straight line might penetrate the interior of the artifact (such as at a bottleneck with significant curvature). Instead, the system uses Dijkstra's algorithm or the heat method to calculate the geodesic displacement modulus between the two points on a triangular mesh. This is the actual path length of the light spot sliding along the curved surface, defined as the surface displacement rate. .

[0065] Simultaneously, in the observation space dimension, the system backtracks the geometric parameters in step S2 to obtain the time. With time The corresponding observation line vector (denoted as) and These two vectors connect the camera optical center at different times to the current surface unit center. The magnitude of the angle between them is calculated using the vector dot product formula:

[0066]

[0067] The calculated result was determined as the rate of change of the observation angle. It is worth mentioning that, to ensure the robustness of the system, a small parallax filtering mechanism is built into the system here: if the calculated... Less than the preset disparity threshold (e.g.) This indicates that the camera is hovering or moving at an extremely slow speed, in which case the calculated ratio is highly unstable. The system will automatically activate the static statistical degradation process, directly using the temporal median of the radiance as the color estimate for that point, instead of forcibly performing subsequent specular mobility calculations.

[0068] Through the above steps, the system successfully deconstructed the complex visual phenomenon into two physical quantities: how far the surface light spot traveled ( And how many degrees the camera rotated () This lays the mathematical foundation for the next step of determining the properties of specular highlights.

[0069] Step S4: Determine the spectrophotometric migration characteristic data of the surface unit based on the ratio of the surface displacement rate to the observation angle change rate;

[0070] In this embodiment, step S4 involves utilizing the physical contradiction between photometric conservation and geometric drift to construct a numerically robust specular migration feature operator.

[0071] Specifically, the system receives two key inputs from step S3: the surface displacement rate, which characterizes the sliding speed of the light spot on the surface of the 3D model. And the rate of change of observation angle, which characterizes the degree of drastic change in observation perspective. In an ideal physical model of diffuse reflection, the inherent texture of an object's surface (such as the blue-and-white pattern on porcelain) is anchored to a geometric grid, and the spatial position of its radiation intensity peak should remain constant regardless of the viewing angle (i.e., Conversely, specular highlights are virtual images determined by the light source, surface normal, and line of sight. When the line of sight is slightly deflected (…), the image will appear as a specular highlight. Highlights will slip significantly on curved surfaces. ).

[0072] Because in actual calculations, when the rate of change of the observation angle... When the value approaches zero (e.g., due to slight camera shake), direct division can lead to numerical singularities or overflow errors. Therefore, this embodiment introduces a preset numerical stability constant into the calculation engine. (For example, the value is) The following robust ratio calculation model was constructed:

[0073]

[0074] in, This refers to the specular migration feature data to be determined. The formula introduces... The regularization term in the denominator ensures that the calculation of eigenvalues ​​remains numerically convergent and smooth even when the disparity is extremely small.

[0075] when When the value is significantly large, it indicates that the surface unit has undergone a huge positional shift under a very small change in viewing angle. This is consistent with the optical drift characteristics of false specular highlights. Therefore, this value is positively correlated with the probability that the surface unit belongs to a false specular highlight.

[0076] when When the value approaches zero, it indicates that the radiation peak at that location is static relative to the surface geometry, consistent with the material anchoring characteristics of the inherent texture.

[0077] Therefore, the surface displacement rate / observation angle transformation rate criterion used in this embodiment essentially constructs an optical kinematic filter independent of the color space, which completely decouples the color attributes of the texture itself. Even white gold-painted or light-colored glazed surfaces that are extremely similar to highlights can be filtered as long as they remain stationary relative to the geometric surface (i.e., the numerator term is not specified). This ratio operator can accurately identify and preserve the inherent texture; conversely, only signals that physically shift with the viewing angle are marked as highlights. This mechanism fundamentally solves the technical problem of false texture confusion, ensuring the integrity of the semantic information on the surface of cultural relics.

[0078] Step S5: Generate a specular confidence mask based on specular migration feature data, and perform weighted decoupling processing on the multidimensional radiation distribution dataset based on the specular confidence mask to generate an initial diffuse texture map and material parameter map;

[0079] In this embodiment, step S5 is a dual-channel signal separation and inversion process based on physical optics. Its core logic lies in using the specular features obtained in the previous steps as prior knowledge to decouple the original radiation data, which mixes diffuse and specular reflections, into independent intrinsic color and material properties.

[0080] Before implementation, to achieve accurate inversion of the microscopic physical properties of the object's surface, the system pre-configures a library of microsurface optical scattering models. This model library preferably adopts the Cook-Torrance BRDF (Bidirectional Reflectance Distribution Function) architecture, whose core assumption is that the object's surface is composed of countless tiny plane mirrors. In actual configuration, the system selects the GGX or Beckmann distribution as the normal distribution function (NDF) because they can be obtained through a single distribution width parameter (i.e., roughness). This model accurately describes the attenuation profile of specular highlights at different viewing angles. The introduction of this physical model allows the system not only to remove highlights but also to extract material information from them.

[0081] During actual operation, the system first reads the specular migration feature data stored on the vertices of the geometric model in step S4. The system uses a pre-defined nonlinear mapping function (such as the Sigmoid function with a dynamic sensitivity parameter) to transform it into a value range within... The normalized probability values ​​between these values ​​are used. To prevent interference from single-point noise, the system utilizes the topological connectivity of the 3D mesh to perform spatial smoothing filtering (such as Laplacian smoothing or Gaussian diffusion) on the probability field. This process eliminates isolated noise points and generates a spatially continuous specular confidence mask, accurately outlining the range of specular areas on the artifact's surface.

[0082] Subsequently, the system initiates parallel dual-channel decoupling processing based on this mask. In the first channel (diffuse texture extraction), the system aims to recover the inherent color of the artifact's surface. For each surface unit, the system constructs a probability value normalized to the specular highlight. Negatively correlated fusion weights (e.g.) Using this weight, the system performs a weighted average calculation of pixel radiance intensity values ​​across all viewpoints in the multidimensional radiance distribution dataset. Under this logic, observation data that is determined to be specular highlights ( Its weight approaches zero, thus it is automatically removed from the fusion result; while diffuse data ( This weighted approach then dominates the final color calculation. This weighting mechanism effectively washes away false bright spots that move with the viewpoint, generating a pure initial diffuse texture map.

[0083] The second channel (material parameter inversion) aims to uncover the physical properties of cultural relics. The system filters out those relics whose normalized probability values ​​in the multidimensional radiation distribution dataset are greater than a preset fitting threshold (e.g., ...). The system uses a subset of data points representing the highlight moments captured by the surface unit at a specific viewpoint. The system combines the observation angle values ​​(input variables) and pixel radiance values ​​(observation variables) from these data points to form an observation sample, which is then fitted to a pre-defined BRDF model using a nonlinear least squares method (such as the Levenberg-Marquardt algorithm). Through iterative optimization, the system inversely solves for the distribution width parameter that minimizes the model error and directly determines it as the roughness parameter of the surface unit. Finally, these parameters are mapped to generate a material parameter map.

[0084] Furthermore, to ensure strict self-consistency of the physical logic, the system performs a roughness-mask mutual exclusion check here. The system compares the generated material parameter map with the specular confidence mask: if the roughness parameter retrieved from a certain surface unit... Greater than the preset diffuse reflection threshold (e.g.) (representing a matte surface), but its corresponding specular confidence normalized probability value Still significantly higher (e.g.) The system determines that the specular detection at this location is a false positive (possibly caused by an illusion resulting from periodic texture movement). In this case, the system forcibly adjusts the mask probability value at this location. Reset to This triggers the first channel to recalculate the diffuse texture of the area, thus completely eliminating the logical paradox.

[0085] Step S6: Determine the reference lighting direction vector in the color space, and construct the difference vector between the initial diffuse texture map and the multi-view image sequence in the color space;

[0086] In this embodiment, step S6 aims to construct a chromaticity mathematical framework based on vector analysis in order to quantify the residual signal between the texture after removing highlights in the preceding steps and the original real observation image.

[0087] Specifically, the system first needs to establish a physical benchmark for measuring color deviation. This embodiment strictly limits all photometric calculations to a standard three-dimensional RGB color space (Euclidean RGB Space). In this space, any color is represented with respect to the origin. A three-dimensional vector originating from a given point. To decouple the illumination components, the system must determine a reference illumination direction vector. Physically, it represents the chromaticity direction of the main light source in the scene.

[0088] In actual configuration, the system employs a dual strategy to determine this vector:

[0089] Prior information mode: If a standard color chart or illuminance meter is placed at the data acquisition site, the system will directly read the color temperature data of the light source (such as a D65 light source) and convert it into a normalized RGB vector as... .

[0090] Adaptive Default Mode (for uncalibrated scenarios): When prior information about the light source is lacking, the system performs geometry construction based on the white light assumption. Specifically, the system defines the zero point of coordinates representing pure black in the RGB color space. ) and the full-scale coordinate point representing pure white (e.g., 8-bit color depth). Subsequently, the system constructs a spatial diagonal vector connecting these two points and normalizes it, using it as the default reference illumination direction vector. This design implicitly assumes that the ambient light is neutral white light (i.e., the R, G, and B components have equal energy), providing a robust initial axis for subsequent orthogonal decomposition.

[0091] After establishing the lighting baseline, the system proceeds to construct the difference vector. This process is essentially a reverse verification experiment. The system utilizes a graphics rendering engine (such as OpenGL or a self-developed rasterizer) to read the initial diffuse texture map generated in step S5 and the camera pose parameters calculated in step S1. Through perspective projection transformation, the system remaps this texture map, with specular highlights removed, back to the imaging plane of each original camera, rendering the corresponding reprojected verification image.

[0092] At this point, for each valid pixel location in the multi-view image sequence, there are two key vector entities in the system:

[0093] Observation pixel vector Extracted from the original acquired image, it contains all the lighting information of the real scene (including diffuse reflection, residual highlights, ambient light, etc.).

[0094] Reprojection pixel vector Extract the self-weight projection verification image, which only contains the diffuse color information currently estimated by the system.

[0095] The system performs vector subtraction operations in the three-dimensional RGB color space:

[0096]

[0097] Calculated difference vector The chromaticity residual between the observed and predicted values ​​was precisely quantified. Physically, this vector contains both incorrectly removed intrinsic texture details (loss due to excessive despeccing) and optical artifacts that were not completely eliminated. The fundamental purpose of constructing this vector is to mathematically separate these two mixed components along the illumination direction using projection geometry in subsequent steps.

[0098] Step S7: Perform orthogonal decomposition on the difference vector based on the reference lighting direction vector, and perform anisotropic compensation on the initial diffuse texture map according to the vertical component obtained from the decomposition to generate the target texture map;

[0099] In this embodiment, step S7 is the key logical closed loop for realizing the removal of false images and the preservation of true images. This step elevates from simple image processing to the level of vector space analysis, aiming to solve a common pain point in traditional highlight removal algorithms—that is, how to distinguish white highlights from white inherent textures (such as white glaze or gold-painted patterns on porcelain).

[0100] In practice, the system first calls the reference illumination direction vector determined in step S6. As the principal axis in the color space, the difference vector corresponding to each surface unit is obtained. Difference vector Essentially, it records all the information loss between the texture after specular removal and the actual observation. In order to accurately recover the inherent texture details that were mistakenly deleted, the system uses the orthogonal projection principle in linear algebra to decouple the difference vector by physical dimension.

[0101] Specifically, the system performs the following orthogonal decomposition calculations in parallel within the GPU shader:

[0102] First, the system calculates the projection of the difference vector onto the reference illumination direction to obtain the parallel component. :

[0103]

[0104] Physically, Represents the color of the light source (usually white or a neutral color), therefore The intensity gain caused by strong light illumination was precisely quantified. This component is the culprit behind specular highlights and represents an optical interference that should be permanently eliminated.

[0105] Next, the system uses vector subtraction to extract the vertical component. :

[0106]

[0107] Physically, Perpendicular to the illumination axis, it means that it does not contain any increments in the brightness dimension, but purely carries the deviation information of chromaticity and saturation. In other words, this component represents the inherent color of the object that was misjudged as a highlight by the preceding algorithm and lost (for example, after a yellow gold-painted spot is mistakenly deleted, its residual will contain a significant yellow component, which will inevitably deviate from the white illumination axis in RGB space).

[0108] After decoupling the physical components, this embodiment further introduces rigorous anisotropic compensation decision logic to prevent the introduction of computational noise:

[0109] The system calculates the vertical component. Length of the module And compare it with a preset color difference tolerance threshold (e.g.) Compare using normalized RGB units:

[0110] like If the difference is less than this threshold, the system determines that the difference is caused only by sensor thermal noise or a slight lighting estimation error, and is therefore an invalid signal. In this case, the system keeps the corresponding pixels of the initial diffuse texture map unchanged and does not perform any compensation, thus ensuring the purity of the image.

[0111] like If the value is greater than or equal to the threshold, the system determines that a significant texture deletion event has occurred (e.g., a specular mask covers a non-spectral light-colored pattern), and the system then performs anisotropic repair.

[0112] During the restoration phase, instead of simply adding the vertical component back to the texture, the system constructs a chroma compensation vector. To smooth the repair boundary, the system dynamically calculates the anisotropic repair coefficient based on the modulus length. (For example, using a linear gradient function), and performing a weighted calculation: .

[0113] Finally, the system extracts the initial texture vector at the corresponding position in the initial diffuse texture map. Perform vector overlay operation:

[0114]

[0115] After performing vector overlay, the system will process the resulting vector. Perform a gamut compliance check. If any RGB component value after overlay exceeds the defined range of the color space (e.g., exceeding 255 in 8-bit depth), the system uses a tone mapping operator (such as the Reinhard operator) to compress the high dynamic range data, or reduces the brightness while keeping the hue unchanged, to ensure that the final output target texture map does not suffer overexposure distortion on standard display devices.

[0116] Through this operation, the system pushes the lost colors back into the texture map along the direction perpendicular to the light in the color space, while resolutely discarding the luminance component parallel to the light direction. This anisotropic processing method ensures that the final generated target texture map completely eliminates glaring highlights (because no additional...). And it perfectly recovered the inherent texture details that had been accidentally damaged (because it was added back). This achieves true high-fidelity digital reconstruction.

[0117] Specifically, this orthogonal decomposition strategy based on the reference illumination direction extracts the vertical component. In essence, this mathematically refines the chromaticity residuals masked by strong light. Because... Orthogonal to the illumination axis, it naturally eliminates the influence of light intensity, retaining only the spectral reflectance characteristics of the object's surface. Therefore, backfilling the texture with the anisotropically weighted vertical component can completely remove specular reflections (parallel components). At the same time, it restores the hue and saturation of the damaged area to the greatest extent, achieving the excellent effect of removing light without losing color.

[0118] It is worth noting that, in order to further eliminate the systematic error caused by the reference illumination direction estimation deviation, this embodiment further executes an illumination vector self-calibration subroutine after completing the above-mentioned preliminary orthogonal decomposition and texture generation. The specific description is as follows:

[0119] Specifically, the system first performs high-confidence sample screening, traversing all surface units, reading the high-brightness confidence mask generated in step S5, and screening out samples whose normalized probability value is greater than the preset high-brightness sample screening threshold (e.g., The surface unit set. These units are considered to be the most defined specular regions in the system, and therefore the optical information they carry best reflects the physical characteristics of the light source.

[0120] Subsequently, the system constructs a set of parallel component features. For each high-confidence surface unit selected above, the system extracts its parallel components that were separated in the initial orthogonal decomposition. At a physical level, these parallel components represent the specular intensity vectors that the algorithm deems should be removed. This is true if the initial reference lighting direction... If it is completely accurate, then all of these... Vectors in a three-dimensional color space should be strictly collinear and point in the direction of the real light source; however, if Due to the deviation, the distribution of these vectors will exhibit a scattering state with the direction of the real light source as the axis.

[0121] To recover the true orientation from this scattering data, the system performs principal component analysis (PCA) on the collected set of parallel components. The specific calculation logic is as follows:

[0122] The system builds a Data matrix ,in To determine the number of samples selected, each row corresponds to one parallel component. The RGB values.

[0123] Calculate the covariance matrix of this matrix. .

[0124] For covariance matrix Perform eigenvalue decomposition to extract the eigenvector corresponding to the largest eigenvalue.

[0125] The direction of the first principal component statistically represents the direction with the largest variance in the dataset, and physically corresponds to the true principal axis of illumination that all specular components point to. The system standardizes this feature vector and determines it as the corrected illumination direction vector. .

[0126] Finally, the system performs deviation verification and iterative decision-making to calculate the corrected value. With the initial reference vector Angle deviation between :

[0127]

[0128] Determine whether the deviation angle is greater than the preset angle deviation tolerance threshold (e.g.) ):

[0129] like If the value is less than or equal to this threshold, the system determines that the initial illumination estimate is accurate enough, and no adjustment is needed; the current texture generation result is directly output.

[0130] like If the value exceeds this threshold, the system determines that there is a significant illumination estimation error. At this point, the calculated... Replace the original This triggers a re-decomposition command. Based on the new illumination vector, the system will re-perform the orthogonal decomposition and anisotropic compensation processing described in step S7 on the difference vector in step S6.

[0131] Through this adaptive closed-loop mechanism, the system can use errors to correct errors and automatically converge to the most realistic lighting parameters in unknown light source environments, ensuring that the final generated target texture map will not produce color cast or residual light spots due to lighting axis deviation.

[0132] Step S8: Render the initial 3D geometric model based on the target texture map and material parameter map to generate a 3D digital reconstruction model and visualize it.

[0133] In this embodiment, step S8 is the final output of the entire digital reconstruction process. Its task is to re-integrate the geometric structure, diffuse color and physical material properties separated in the previous steps to construct a relightable 3D digital asset with lighting interaction capabilities.

[0134] Specifically, to achieve high-fidelity visual presentation, this embodiment builds a real-time visualization engine based on the Physically Based Rendering (PBR) standard. The system loads an initial 3D geometric model as a skeleton and maps the target texture map generated in step S7 to the model's diffuse / albedo channel according to texture coordinates (UV coordinates). This texture map, having undergone rigorous orthogonal decomposition and anisotropic compensation, removes fixed highlights from the original shooting environment, revealing the pure inherent color of the artifact. Simultaneously, the system maps the material parameter map obtained inverted in step S5 to the roughness channel and metallic channel of the rendering pipeline. This mapping gives the digital model texture, transforming it from a flat, colored model into a physically realizable entity that correctly responds to virtual lighting based on surface microstructures (such as the smoothness of glaze or the roughness of rust).

[0135] To verify the reconstruction results and provide a visual demonstration, the system constructs a dynamic virtual lighting environment in the rendered scene. This embodiment employs Image Based Lighting (IBL) technology, utilizing High Dynamic Range Environment Mapping (HDRi) to simulate a museum exhibition hall or natural light environment. The system calls the rendering shader to calculate the interaction between virtual light and the model surface in real time.

[0136] At this point, for areas with low roughness parameters on the model (such as the inverted porcelain glaze), the rendering engine generates sharp and bright virtual specular reflections based on the Fresnel effect; while for areas with high roughness parameters (such as the ceramic base), light is scattered, resulting in a matte texture. Most importantly, when the user rotates the model or moves the virtual light source through the interactive interface, these virtual highlights, calculated in real-time by the rendering engine, create smooth and natural dynamic transitions on the model's surface, completely replicating the optical properties of the real physical world, while the rigid, fixed highlights of the original shooting are no longer present.

[0137] Finally, the system outputs the rendered 3D scene to a high-resolution display terminal and supports exporting the reconstructed model to a common format (such as glTF or USDz). These formats fully encapsulate the mesh, target texture, and material parameters, forming a highly reusable 3D digital archive of cultural heritage that can be directly applied to virtual museum displays, artifact restoration simulations, or cross-platform academic exchanges.

[0138] In summary, this embodiment constructs a highly adaptive digital reconstruction system by introducing quality perception based on frequency domain entropy at the front end, building a motion differential operator based on manifold geometry in the middle, and performing texture compensation based on vector orthogonal decomposition and illumination self-calibration closed loop at the back end. This system not only effectively overcomes optical drift interference caused by the complex optical properties of artifact surfaces, but also ensures stable operation under sensor noise, small parallax, and unknown light source environments through logical mutual exclusion checks at the physical level and robust design at the numerical level. The final output digital model significantly outperforms reconstruction schemes based on traditional diffuse reflection assumptions in both geometric accuracy and texture fidelity.

[0139] Example 2:

[0140] As shown in Figure 4, the cultural heritage 3D digital reconstruction and display system includes:

[0141] The data acquisition module is used to acquire the initial three-dimensional geometric model of the target object to be reconstructed and the corresponding multi-view image sequence;

[0142] The radiation distribution construction module is used to map pixel data in multi-view image sequences to surface units of an initial three-dimensional geometric model, and construct a multi-dimensional radiation distribution dataset containing the observation angle dimension and the radiation intensity dimension.

[0143] The rate calculation module is used to calculate the surface displacement rate of the radiation intensity peak of the surface unit and the corresponding observation angle transformation rate based on the multidimensional radiation distribution dataset.

[0144] The specular feature determination module is used to determine the specular migration feature data of surface units based on the ratio of surface displacement rate to observation angle change rate.

[0145] The texture decoupling module is used to generate a specular confidence mask based on specular migration feature data, and to perform weighted decoupling processing on the multidimensional radiation distribution dataset based on the specular confidence mask to generate an initial diffuse texture map and a material parameter map.

[0146] The lighting analysis module is used to determine the reference lighting direction vector in the color space and construct the difference vector between the initial diffuse texture map and the multi-view image sequence in the color space.

[0147] The texture compensation module is used to perform orthogonal decomposition on the difference vector based on the reference lighting direction vector, and to perform anisotropic compensation on the initial diffuse texture map according to the vertical component obtained by decomposition to generate the target texture map.

[0148] The rendering and display module is used to render the initial 3D geometric model based on the target texture map and material parameter map, generate a 3D digital reconstruction model, and display it visually.

[0149] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0150] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0151] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for three-dimensional digital reconstruction and display of cultural heritage, characterized in that, Includes the following steps: Obtain the initial 3D geometric model of the target object to be reconstructed and the corresponding multi-view image sequence; map the pixel data in the multi-view image sequence to the surface units of the initial 3D geometric model to construct a multi-dimensional radiation distribution dataset containing the observation angle dimension and the radiation intensity dimension; based on the multi-dimensional radiation distribution dataset, calculate the surface displacement rate of the radiation intensity peak of the surface unit and the corresponding observation angle transformation rate; determine the spectrophotometric migration feature data of the surface unit according to the ratio of the surface displacement rate to the observation angle transformation rate. A high-brightness confidence mask is generated based on the high-brightness migration feature data, and a weighted decoupling process is performed on the multidimensional radiation distribution dataset based on the high-brightness confidence mask to generate an initial diffuse texture map and a material parameter map. A reference illumination direction vector is determined in the color space to which the initial diffuse texture map belongs, and a difference vector between the initial diffuse texture map and the multi-view image sequence in the color space is constructed; orthogonal decomposition is performed on the difference vector based on the reference illumination direction vector, and anisotropic compensation is performed on the initial diffuse texture map according to the vertical component obtained by decomposition to generate a target texture map; The initial 3D geometric model is rendered based on the target texture map and the material parameter map to generate a 3D digital reconstruction model and display it visually.

2. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 1, characterized in that: The construction process of the multidimensional radiation distribution dataset includes: for each surface unit on the initial three-dimensional geometric model, based on the camera pose parameters corresponding to the multi-viewpoint image sequence, selecting an effective set of cameras that can visually cover the surface unit; determining the geometric center coordinates of the surface unit, and projecting the geometric center coordinates inversely onto the imaging plane of each camera in the effective camera set to obtain the corresponding pixel radiation intensity value; constructing an observation line vector connecting the optical center of each effective camera with the geometric center coordinates of the surface unit; calculating the cosine value of the angle between the observation line vector and the normal vector of the surface unit, and converting the cosine value into an observation angle value; forming a binary sequence of the observation angle value and the pixel radiation intensity value to construct the multidimensional radiation distribution dataset corresponding to the surface unit.

3. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 2, characterized in that: Based on the multidimensional radiation distribution dataset, the surface displacement rate of the radiation intensity peak of the surface unit and the corresponding observation angle transformation rate are calculated, including: determining the topological neighborhood range of the surface unit on the initial three-dimensional geometric model; in the multidimensional radiation distribution dataset, according to the time sequence of observation times, sequentially identifying the surface location of the radiation intensity peak within the topological neighborhood range, and constructing the peak movement path; calculating the geodesic displacement modulus of the peak movement path between adjacent observation times to obtain the surface displacement rate; obtaining the observation line vector corresponding to adjacent observation times, calculating the angle modulus between the observation line vectors, and determining the angle modulus as the observation angle transformation rate.

4. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 3, characterized in that: The generation process of the hyperbola confidence mask includes: using a nonlinear mapping function to map the hyperbola migration feature data into a normalized probability value with a value range between 0 and 1; obtaining the mesh topology connection relationship of the initial three-dimensional geometric model, and performing spatial smoothing filtering on the normalized probability value based on the mesh topology connection relationship to generate a hyperbola confidence mask with spatial continuity at the surface unit.

5. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 4, characterized in that: Based on the high-brightness confidence mask, a weighted decoupling process is performed on the multidimensional radiation distribution dataset to generate an initial diffuse texture map and a material parameter map. This includes: extracting the normalized probability value corresponding to the surface unit in the high-brightness confidence mask; constructing a fusion weight negatively correlated with the normalized probability value; using the fusion weight to perform a weighted average calculation on the pixel radiation intensity values ​​in the multidimensional radiation distribution dataset to obtain the texture color of the initial diffuse texture map; selecting a subset of data in the multidimensional radiation distribution dataset whose normalized probability value is greater than a preset fitting threshold; using the observation angle value and pixel radiation intensity value in the subset of data as observation samples; using the least squares method to fit the observation samples to a preset micro-surface bidirectional reflection distribution function model; solving the distribution width parameter in the micro-surface bidirectional reflection distribution function model in reverse; and determining the solved distribution width parameter as the roughness parameter in the material parameter map.

6. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 5, characterized in that: The color space is a three-dimensional RGB color space; The process of determining the reference illumination direction vector includes: obtaining the prior color information of the light source in the shooting scene, constructing a normalized RGB vector, and determining the RGB vector as the reference illumination direction vector; When prior color information of the light source is lacking, in the three-dimensional RGB color space, the coordinate zero point representing pure black and the full-scale coordinate point representing pure white are determined; a diagonal vector connecting the coordinate zero point and the full-scale coordinate point is constructed, and the diagonal vector is determined as the reference illumination direction vector.

7. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 6, characterized in that: The process of constructing the difference vector includes: based on the camera pose parameters corresponding to the multi-view image sequence, using a graphics rendering engine to map the initial diffuse texture map back to the imaging plane of the multi-view image sequence to generate a corresponding reprojection verification image; in the three-dimensional RGB color space, extracting the RGB values ​​of the pixels at the corresponding positions in the multi-view image sequence as the observed pixel vector, and extracting the RGB values ​​of the pixels at the corresponding positions in the reprojection verification image as the reprojection pixel vector; calculating the difference vector by subtracting the reprojection pixel vector from the observed pixel vector.

8. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 7, characterized in that: The generation process of the target texture map includes: performing orthogonal decomposition on the difference vector based on the reference lighting direction vector to obtain parallel and perpendicular components, the calculation formula of which is: in, Represents the difference vector. Represents the reference lighting direction vector. This represents the parallel component that is parallel to the reference illumination direction vector. The vertical component is represented by a vector perpendicular to the reference illumination direction. The magnitude of the vertical component is calculated, and it is determined whether the magnitude is less than a preset chromatic aberration tolerance threshold. If it is, the initial diffuse texture map is directly determined as the target texture map. Otherwise, an anisotropic repair coefficient positively correlated with the magnitude is determined, and the vertical component is weighted according to the anisotropic repair coefficient to obtain a chromaticity compensation vector. The texture color at the corresponding position in the initial diffuse texture map is extracted, and the RGB value of the texture color is determined as the initial texture vector. The chromaticity compensation vector and the initial texture vector are vector-superimposed to obtain the target texture map.

9. The method for three-dimensional digital reconstruction and display of cultural heritage according to claim 8, characterized in that: After performing orthogonal decomposition on the difference vector based on the reference illumination direction vector, the method further includes: selecting surface units in the high-brightness confidence mask whose normalized probability value is greater than a preset calibration threshold; obtaining the parallel components obtained by the surface units in the orthogonal decomposition process and constructing a set of parallel components; performing principal component analysis on the set of parallel components and extracting the direction of the first principal component as the corrected illumination direction vector; calculating the deviation angle between the corrected illumination direction vector and the reference illumination direction vector; if the deviation angle is greater than a preset deviation tolerance threshold, replacing the reference illumination direction vector with the corrected illumination direction vector and re-performing the orthogonal decomposition on the difference vector.

10. A three-dimensional digital reconstruction and display system for cultural heritage, characterized in that, include: The data acquisition module is used to acquire the initial three-dimensional geometric model of the target object to be reconstructed and the corresponding multi-view image sequence; A radiation distribution construction module is used to map pixel data in the multi-view image sequence to surface units of the initial three-dimensional geometric model to construct a multidimensional radiation distribution dataset containing observation angle dimension and radiation intensity dimension; a rate calculation module is used to calculate the surface displacement rate of the radiation intensity peak of the surface unit and the corresponding observation angle transformation rate based on the multidimensional radiation distribution dataset. The specular feature determination module is used to determine the specular migration feature data of the surface unit based on the ratio of the surface displacement rate to the observation angle transformation rate; the texture decoupling module is used to generate a specular confidence mask based on the specular migration feature data, and perform weighted decoupling processing on the multidimensional radiation distribution dataset based on the specular confidence mask to generate an initial diffuse texture map and a material parameter map; The illumination analysis module is used to determine the reference illumination direction vector in the color space and construct the difference vector between the initial diffuse texture map and the multi-viewpoint image sequence in the color space; the texture compensation module is used to perform orthogonal decomposition processing on the difference vector based on the reference illumination direction vector, and perform anisotropic compensation on the initial diffuse texture map according to the vertical component obtained by decomposition to generate the target texture map. The rendering and display module is used to render the initial three-dimensional geometric model based on the target texture map and the material parameter map, generate a three-dimensional digital reconstruction model, and perform visualization display.

Images