A collaborative processing method for non-standard object surface reduction optimization and texture baking
By using a weighted edge collapse cost function and a closed-loop collaborative mechanism, UV stretching and texture quality are detected in real time, solving the problem of irrecoverable texture damage during the reduction of surfaces in non-standard object 3D simulation models, and achieving model lightweighting and texture fidelity preservation.
Patent Information
- Application Number
- CN202611328070.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-17
AI Technical Summary
When processing non-standard 3D realistic models, existing technologies often lose details due to the reduction of polygons. Furthermore, the separate processing of polygon reduction and texture baking makes it difficult to repair texture distortion. It is also impossible to promptly identify issues such as UV stretching and texture offset during the polygon reduction process, resulting in uncontrollable texture fidelity and irreversible damage.
A weighted edge collapse cost function is used to pre-constrain texture error. Combined with UV stretching detection and texture baking quality detection, a closed-loop collaborative mechanism of reduction-detection-feedback is formed. The reduction process is adjusted in real time through parameter adjustment or local rollback operation to achieve texture fidelity and model lightweighting.
During the reduction process, UV stretching and texture quality are detected in real time. The parameters of the collapse cost function are adjusted through closed-loop feedback to avoid texture damage, thereby achieving model lightweighting while preserving texture fidelity and solving the problem of irreversible texture damage.
Smart Images

Figure CN123118908A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of lightweight 3D simulation model technology, and in particular to a method for co-processing non-standard object surface reduction optimization and texture baking. Background Technology
[0002] Non-standard objects, such as unusual stones, strangely shaped sculptures, and irregularly shaped collectibles, typically feature irregular shapes, continuously changing curved surfaces, complex local concavities and convexities, and rich surface texture details. During the 3D digitization process, these objects usually generate high-polygon, high-texture-resolution 3D realistic models.
[0003] While high-precision 3D simulation models can preserve the original appearance well, they can easily cause problems such as slow loading speed, high rendering pressure and high storage consumption in real-time applications such as mobile display, web page loading, VR / AR interaction and digital exhibition. Therefore, polygon reduction optimization and texture baking are required.
[0004] The existing solutions mainly have the following problems: First, polygon reduction methods that primarily rely on geometric errors are prone to losing details. Existing polygon reduction methods mostly depend on geometric errors, using edge collapse, quadratic error measurement, and other methods to reduce the number of faces in the model. When processing non-standard 3D realistic models, these methods are prone to causing excessive compression of details such as local bumps, hole edges, natural textures, and key display areas.
[0005] Second, the step-by-step processing of polygon reduction and texture baking makes it difficult to repair texture distortion. Existing polygon reduction and texture baking usually adopt a step-by-step processing method, that is, reducing polygons first and then baking. This method cannot timely identify problems such as UV stretching, texture offset, local blurring and edge misalignment during the polygon reduction process. As a result, texture baking can only passively repair based on the polygon reduction result, making it difficult to balance model lightweighting and texture fidelity. Summary of the Invention
[0006] The purpose of this application is to provide a collaborative processing method for non-standard object surface reduction optimization and texture baking, so as to achieve texture fidelity while realizing model lightweighting.
[0007] To achieve the above objectives, this application provides the following solution.
[0008] Firstly, this application provides a method for co-processing non-standard object surface reduction optimization and texture baking, including: The original 3D simulation model of the non-standard object is used as the 3D simulation model before the first round of face reduction, and the value of the face reduction round n is set to 1. The collapse cost of different local regions of the 3D simulation model before the nth round of reduction is calculated using the weighted edge collapse cost function. Based on the collapse cost of different local regions, the 3D simulation model before the nth round of reduction is reduced in terms of surface area to obtain the 3D simulation model after the nth round of reduction in surface area. UV stretching detection and texture baking quality detection are performed on the 3D simulation model after the nth round of surface reduction to obtain the detection results of different local areas; When the detection results of one or more local regions do not meet the surface reduction quality requirements, a parameter adjustment operation or a local rollback operation is performed. The 3D simulation model before the nth round of surface reduction or the 3D simulation model after the local rollback operation is used as the new 3D simulation model before the nth round of surface reduction, and the process returns to the step of "calculating the collapse cost of different local regions of the 3D simulation model before the nth round of surface reduction using the weighted edge collapse cost function". When the detection results of each local region meet the surface reduction quality requirements but do not meet the preset output conditions, the 3D simulation model after surface reduction in the nth round is used as the 3D simulation model before surface reduction in the (n+1)th round. The value of n is increased by 1, and the process returns to the step of "calculating the collapse cost of different local regions of the 3D simulation model before surface reduction in the nth round using the weighted edge collapse cost function". When the detection results of each local area meet the surface reduction quality requirements and reach the preset output conditions, the final surface reduction 3D simulation model is output as the lightweight 3D simulation model.
[0009] Optionally, the weighted edge collapse cost function is: ; in, For local areas The cost of collapse For local areas The geometric error term, For local areas Protection weight penalty item, , For local areas Geometric key weights, For local areas Texture detail weights For local areas The display attention weight, , , These are the weighting coefficients. For local areas Texture mapping variation terms, , This is the adjustment coefficient.
[0010] Optionally, the geometric key weights, texture detail weights, and display attention weights of each local region are obtained by performing three-dimensional feature analysis on the original 3D simulation model; The three-dimensional features include vertex curvature, texture image color gradient, and spatial observation constraint parameters.
[0011] Optionally, the detection results include: a combination of UV stretching, texture error, and texture entropy; When the total UV stretching amount is not greater than the UV protection threshold, the texture error is not greater than the texture error threshold, and the texture entropy is not greater than the texture entropy threshold, the detection result is determined to meet the surface reduction quality requirements; otherwise, the detection result is determined not to meet the surface reduction quality requirements.
[0012] Optionally, the formula for calculating the comprehensive UV stretching amount is: ; ; ; ; in, The total UV stretching amount for local region e. , , These represent the UV area stretching, UV side length stretching, and UV angular distortion rate of the local region e, respectively. , To reduce the UV area of the local region e before and after the front, , To reduce the UV edge length of the local region e before and after the front, , To reduce the UV angle of the local area e before and after the front.
[0013] Optionally, local areas where the detection results do not meet preset conditions are designated as target local areas. For each target local area, the parameter adjustment operation or local rollback operation specifically includes: When the total UV stretching of the target local area is greater than the UV protection threshold, and the texture error is not greater than the texture error threshold, and the texture entropy is not greater than the texture entropy threshold, then the first parameter adjustment operation is performed on the target local area. If the texture error of the target local region is greater than the texture error threshold and the texture entropy is not greater than the texture entropy threshold, then the second parameter adjustment operation is performed on the target local region. If the texture entropy of the target local region is greater than the texture entropy threshold, a local rollback operation is performed on the target local region.
[0014] Optionally, the first parameter adjustment operation includes: Increase the protection weight penalty for the target local area; Or reduce the step size of the reduction area; Alternatively, UV remapping can be performed on a local area of the target.
[0015] Optionally, the second parameter adjustment operation is as follows: Increase the protection weight penalty term for the target local area and decrease the reduction step size.
[0016] Optionally, the partial rollback operation includes: In the nth round of reduction, the vertex positions, UV coordinates, mesh topology, and protection weight penalty terms of the target local region in the 3D simulation model are adjusted sequentially to restore the vertex positions, UV coordinates, mesh topology, and protection weight penalty terms of the target local region to their original state, and the UV protection threshold of the target local region is increased.
[0017] Optionally, the preset output conditions include: The face reduction ratio reaches the target value or the number of face reduction cycles reaches the upper limit.
[0018] According to the specific embodiments provided in this application, this application has the following technical effects.
[0019] This application provides a collaborative processing method for non-standard object polygon reduction optimization and texture baking. During the polygon reduction process, a weighted edge collapse cost function is used to pre-constrain texture errors, reducing the probability of texture damage. After each round of polygon reduction, UV stretching detection and texture baking quality detection are performed. Based on the detection results, the polygon reduction process is adjusted in real time (i.e., parameter adjustment or local rollback), forming a closed-loop collaborative mechanism of polygon reduction-detection-feedback. This avoids over-compression and partially restores texture damage during the polygon reduction process, preventing texture distortion. This application achieves both model lightweighting and texture fidelity. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a flowchart illustrating a method for co-processing non-standard object surface reduction optimization and texture baking, provided in an embodiment of this application.
[0022] Figure 2 This is a schematic diagram illustrating the principle of a non-standard object surface reduction optimization and texture baking collaborative processing method provided in an embodiment of this application.
[0023] Figure 3 This is a schematic diagram of three-dimensional feature analysis provided for an embodiment of this application.
[0024] Figure 4 This is a schematic diagram of the structure of a non-standard object surface reduction optimization and texture baking collaborative processing system provided in an embodiment of this application. Detailed Implementation
[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0026] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0027] Research indicates that current facet reduction techniques lack closed-loop feedback for UV stretching. The absence of a separate UV stretching detection step during facet reduction prevents timely detection and adjustment of mapping deformation after each round of facet reduction. This results in UV stretching issues being discovered only after facet reduction is complete, leading to high repair costs and limited effectiveness. Furthermore, facet reduction methods lack mechanisms to prevent irreversible damage. When edge collapse during facet reduction causes severe loss of local texture information, existing methods cannot determine if the damage has reached an irreversible level, nor can they undo the already executed collapse operation; they can only passively accept the quality loss after facet reduction. The lack of texture entropy constraints and local rollback mechanisms means that once irreversible texture damage occurs during facet reduction, it cannot be recovered.
[0028] Therefore, it is necessary to propose a collaborative processing method for surface reduction optimization and texture baking of non-standard object 3D simulation models. By incorporating the texture mapping change term into the collapse cost function to achieve pre-constraint of texture error, and by implementing a three-level progressive protection mechanism for irreversible damage avoidance, texture damage during the surface reduction process can be partially recovered, thus solving the problems of uncontrollable texture fidelity and unrecoverable damage in existing methods.
[0029] The collaborative processing in this application is reflected in the simultaneous UV stretching judgment and texture baking quality detection during the face reduction process. The parameters of the collapse cost function are adjusted in real time through closed-loop feedback, forming a closed-loop collaborative mechanism of face reduction-detection-feedback, so as to achieve synchronous optimization of face reduction and texture fidelity.
[0030] In one exemplary embodiment, a method for co-processing non-standard object surface reduction optimization and texture baking is provided, such as... Figure 1 As shown, it includes the following steps 101-107.
[0031] Step 101: Use the original 3D simulation model of the non-standard object as the 3D simulation model before the first round of face reduction, and set the value of the face reduction round n to 1. Step 102: Calculate the collapse cost of different local regions of the 3D simulation model before the nth round of reduction using the weighted edge collapse cost function; Step 103: Reduce the number of faces in the 3D simulation model before the nth round of face reduction according to the collapse cost of different local regions, and obtain the 3D simulation model after face reduction in the nth round. Step 104: Perform UV stretching detection and texture baking quality detection on the 3D simulation model after the nth round of surface reduction to obtain the detection results of different local areas; Step 105: When the detection results of one or more local regions do not meet the surface reduction quality requirements, perform parameter adjustment operation or local rollback operation, take the 3D simulation model before the nth round of surface reduction or the 3D simulation model after the local rollback operation as the new 3D simulation model before the nth round of surface reduction, and return to the step of "calculating the collapse cost of different local regions of the 3D simulation model before the nth round of surface reduction using the weighted edge collapse cost function". Step 106: When the detection results of each local region meet the surface reduction quality requirements and do not meet the preset output conditions, the 3D simulation model after surface reduction in the nth round is used as the 3D simulation model before surface reduction in the (n+1)th round. The value of n is increased by 1, and the process returns to the step of "calculating the collapse cost of different local regions of the 3D simulation model before surface reduction in the nth round using the weighted edge collapse cost function". Step 107: When the detection results of each local area meet the surface reduction quality requirements and reach the preset output conditions, output the final surface reduction 3D simulation model as the lightweight 3D simulation model.
[0032] The non-standard object reduction optimization and texture baking collaborative processing method in the above embodiments is an edge collapse optimization method based on texture error constraints. Its core feature is the collaborative processing of reduction and texture baking—not a step-by-step process of reducing surface area first and then baking, but rather simultaneous texture quality monitoring and feedback adjustment during the reduction process, forming a closed-loop collaborative mechanism of reduction-detection-feedback. The execution subject of the above method embodiments is a computer / server or other hardware device.
[0033] In another exemplary embodiment, the above-mentioned closed-loop collaborative mechanism of reduction-detection-feedback specifically includes: real-time detection of UV stretching and texture quality during the reduction process; differential feedback of detection results to the reduction strategy; triggering local rollback when texture damage reaches the irreversible threshold to restore the original state of the region; and dynamically adjusting the parameters of the region after rollback to prevent the same type of damage from recurring.
[0034] In another exemplary embodiment, the aforementioned non-standard objects refer to solid objects with irregular shapes, continuously changing curved surfaces, and complex surface textures, such as rare stones, irregularly shaped sculptures, and irregular collectibles. The essential difference between non-standard objects and standardized industrial products lies in the fact that standardized industrial products (such as mechanical parts, electronic components, and furniture accessories) can be parametrically designed using two-dimensional drawings or three-dimensional models and mass-produced using standardized processes such as molds and machine tools; their geometry can be precisely described by mathematical formulas. Non-standard objects, however, cannot be designed using drawings; their geometric and textural information can only be directly obtained through 3D scanning equipment. Their shapes are formed naturally or artificially sculpted, exhibiting high complexity and irregularity. Therefore, the realistic modeling of non-standard objects differs fundamentally from the modeling of standard industrial products: the former can only "reproduce" (faithfully record the original form), while the latter can be "designed" (generate a model from drawings). The method described in this application is specifically proposed to address the surface reduction optimization needs of this special modeling object, the non-standard object.
[0035] In another exemplary embodiment, the 3D simulation model described above can be a triangular mesh model or a mesh model formed by surface reconstruction of point cloud data; the noise reduction in the preprocessing can be done by statistical filtering, radius filtering or bilateral filtering, and the hole filling can be done by Poisson reconstruction or the forward front method.
[0036] In another exemplary embodiment, such as Figure 2 As shown, step 101 above can be implemented using the following step 1.
[0037] Step 1: Obtain the original 3D simulation model and original texture image. A high-precision 3D simulation model and corresponding texture image of the non-standard object to be processed are obtained. The 3D simulation model is preprocessed, including denoising, hole filling, normal vector calculation, curvature calculation, scale normalization, and establishing a mapping relationship between the surface of the 3D simulation model and the texture image. The preprocessed 3D simulation model is used as the first 3D simulation model before surface reduction. It should be noted that in this embodiment, the surface reduction execution and detection feedback form a closed-loop collaborative relationship, rather than being processed step-by-step. Surface reduction is not completed all at once, but is gradually advanced in a detection-feedback-adjustment loop until the output conditions are met or a local rollback is triggered.
[0038] In another exemplary embodiment, such as Figure 2 As shown, step 102 above can be implemented using steps 2-4 as follows.
[0039] Step 2: Three-Dimensional Feature Analysis. The purpose of this step is to simultaneously analyze the surface features of the model from three different technical dimensions, providing a basis for subsequent differential surface reduction. It should be noted that curvature analysis, Sobel gradient, texture entropy, and QEM are all existing technologies. The combination and synergistic relationship in this embodiment are not simple feature combinations, but rather the synergistic effect of the analysis results from the three dimensions in subsequent steps. The protection requirements of the three dimensions come from three different technical fields: geometric structure, image processing, and human-computer interaction. Figure 3 As shown, features are extracted from different technical dimensions to improve the completeness of the protection weight expression.
[0040] Geometric critical region analysis—identifying high curvature regions by calculating the curvature values of mesh vertices to locate locations with drastic shape changes. For example, the normalized mean curvature threshold is set to 0.7 (low value [0,0.2], medium value [0.2,0.5], high value [0.5,1], and in this application embodiment, a high value of 0.5 or higher is recommended to better protect geometric details. 0.7 is located in the last 30% quantile, which can cover vertices with drastic shape changes while avoiding mislabeling flat areas), preventing these regions from being over-compressed. Texture detail region analysis—identifying areas with dense natural textures, locating dense textures by analyzing the color gradient of the texture image, for example, setting the color gradient amplitude threshold to 30 (Sobel gradient amplitude (low value [0,15], medium value [15,45], high value [45,255], this application embodiment recommends using the medium value 30 to balance detail preservation and computational efficiency) ranges from 0 to 255, 30 corresponds to a lower gradient level, which can effectively filter out flat texture areas while retaining areas with obvious texture changes), to prevent texture blurring after faceting; Display Attention Area Analysis—Based on the spatial constraints of a 3D display scene, this method determines the areas where users are most likely to observe and interact. Unlike existing 2D image saliency detection methods, the input to this embodiment is the spatial observation constraint parameters of the 3D display scene (including viewpoint position, interaction distance, and scene layout), rather than 2D image pixel features. All three dimensions are indispensable: focusing solely on geometry will miss texture, focusing solely on texture will miss structure, and ignoring display attention will remove the parts that users care about most.
[0041] In addition to curvature and normal vector changes, the aforementioned key geometric regions can also be identified using geometric features such as local shape diameter functions, heat kernel signatures, or geodesic distances; texture detail regions can be identified using Gabor filter responses, local binary patterns, or wavelet transform coefficients, in addition to color gradients and edge densities; and display areas of interest can be determined using eye-tracking data, user dwell time statistics, or deep learning-based attention prediction models, in addition to 3D display scene spatial constraint parameters.
[0042] Step 3: 3D Fusion to Generate a Multi-Layer Protection Weight Map. The purpose of this step is to synthesize the analysis results of the three dimensions in Step 2 into a unified "protection map". The geometric key region, texture detail region, and display concern region are fused to generate a multi-layer protection weight map of the 3D realistic model surface, including protection weights at three granularities: region level, vertex level, and edge level. The protection weight W = αG + βT + γV, where G is the geometric key weight, T is the texture detail weight, V is the display concern weight, and α, β, and γ are weight coefficients. It should be noted that the fusion method of W = αG + βT + γV is not a simple weighted average superposition—the three input weights G, T, and V come from different technical fields and have different physical dimensions and numerical ranges. The minor innovation of this embodiment lies in assigning a fixed value / range to the commonly used formula, making it more optimal in this scenario. Specifically, before fusion, G, T, and V are normalized to unify the numerical range of the three dimensions to the [0,1] interval, and then the dimensional balance is achieved through the weighted summation of α, β, and γ. Recommended values: α∈[0.3,0.5] (geometric dimension weight), β∈[0.3,0.5] (texture dimension weight), γ∈[0.2,0.3] (display dimension weight), and α+β+γ=1, ensuring the sum of the weights of the three dimensions is 1, achieving normalized weighted fusion. The weight coefficients α, β, and γ simultaneously serve the dual purpose of inter-domain normalization and inter-dimensional balance. Regions with high weights are less likely to be compressed during face reduction, while regions with low weights are given priority for face reduction.
[0043] This 3D fusion weighted protection process simultaneously analyzes the model's surface features from three different technical dimensions: geometrically critical regions, textured detail regions, and display interest regions based on the spatial constraints of the 3D display scene. It then fuses these features to generate a multi-layered protection weight map W=αG+βT+γV, achieving differentiated surface reduction. The protection requirements in these three dimensions originate from three different technical fields: geometric structure, image processing, and human-computer interaction. Features are extracted from each technical dimension to enhance the completeness of the protection weight representation.
[0044] In addition to linear weighting, the above-mentioned protection weights can also be fused using nonlinear fusion methods such as attention weighting, feature concatenation followed by mapping through a fully connected layer, or learnable fusion networks; the weight coefficients α, β, and γ can be set manually or can be adaptively learned through training data.
[0045] Step 4: Weighted Edge Collapse and Surface Reduction. A weighted edge collapse cost function is constructed based on the multi-layered protection weight graph, and a round of surface reduction is performed on the 3D simulation model. The weighted edge collapse cost function is: ,in, For local areas The cost of collapse For local areas The geometric error term, For local areas The protection weight penalty term here refers to the protection weight mentioned in step 3 above. , For local areas Geometric key weights, For local areas Texture detail weights For local areas The display attention weight, , , These are the weighting coefficients. For local areas Texture mapping variation terms, , This is the adjustment coefficient.
[0046] Traditional cost functions only contain geometric errors Texture errors do not affect the collapse decision, leading to uncontrollable compression of texture details during the facet reduction process. This application's embodiments introduce a texture mapping variation term. Texture error is pre-constrained into the collapse decision: Based on the calculation of changes in the texture mapping after collapse, edges with larger changes in texture map have higher collapse costs and are ranked lower, thus reducing the probability of texture damage. This makes edges with large texture changes less likely to be collapsed first. Texture error plays a role in the collapse decision-making stage, enabling edge collapse optimization based on texture error constraints. The protection weight penalty term Wp(e) allows the 3D fusion protection weight to directly influence the collapse order through the cost function. Three cost terms... , and The collapse order is determined collaboratively: Cg(e) provides the geometric accuracy baseline, and Wp(e) superimposes the protective weight constraints. Further texture quality constraints are superimposed, and the three elements form a dynamic balance through the adjustment of λ1 and λ2.
[0047] In addition to the weighted edge collapse cost function mentioned above, the geometric error term in the collapse cost calculation function can use quadratic error metric (QEM), Hausdorff distance, or volume preservation error; the texture mapping variation term... It can be calculated using texture coordinate offset, pixel color difference, or texture gradient change.
[0048] In another exemplary embodiment, the above detection results include: combining UV stretching amount, texture error, and texture entropy. This embodiment incorporates both UV stretching detection and texture quality detection into a facet reduction loop to form differentiated closed-loop feedback. Different feedback strategies are used for different types of errors in each detection. Step 104 can be implemented using steps 5 and 6 as follows.
[0049] Step 5: UV stretching detection. The purpose of this step is to detect whether the texture mapping has been stretched after each round of face reduction, which is the first detection in the closed-loop feedback. The UV stretching amount is determined by comparing the degree of UV mapping deformation in the corresponding local areas before and after face reduction. The specific calculation method is as follows: (1) UV area stretching amount: The degree of stretching is measured by comparing the rate of change of UV area in local areas before and after the reduction; (2) UV side length stretching: The degree of stretching is measured by comparing the length change rate of the corresponding UV edges before and after the reduction front; (3) UV angle distortion rate: The degree of distortion is measured by comparing the changes in UV angles before and after the reduction; (4) Comprehensive UV stretching: .
[0050] in, The total UV stretching amount for local region e. , , These represent the UV area stretching, UV side length stretching, and UV angular distortion rate of the local region e, respectively. , To reduce the UV area of the local region e before and after the front, , To reduce the UV edge length of the local region e before and after the front, , To reduce the UV angles of the local region e before and after the face reduction, w1, w2, and w3 are weighting coefficients, satisfying w1 + w2 + w3 = 1. When the total UV stretching amount (recommended thresholds: low [0, 0.1], medium [0.1, 0.3], high [0.3, 1], with the recommended medium value of 0.2 in this application) exceeds the current UV protection threshold, the region is marked as abnormal, and feedback adjustment is triggered: increasing the protection weight of the region, reducing the face reduction step size in the next round, increasing the texture sampling accuracy, or performing local UV remapping.
[0051] In addition to using UV area change rate, UV side length change rate and UV angle distortion rate, the above-mentioned comprehensive UV stretching amount can also use indicators such as texture distortion energy, conformal mapping error or spring system deformation amount; the UV stretching threshold can be set to a fixed value or can be adaptively adjusted according to the local features of the model.
[0052] Step 6: Texture baking quality detection. The purpose of this step is to further detect whether the texture has shifted, blurred or misaligned edges. It is the second detection in the closed-loop feedback. At the same time, this application embodiment introduces texture entropy constraint in texture baking quality detection - for each local area of the 3D simulation model surface, calculate the color information entropy of its texture image. The lower the texture entropy, the less texture information in the area. The specific calculation method is as follows: (1) Texture shift detection: calculate the pixel-level color difference of the texture image before and after the face reduction, and measure the degree of shift by pixel distance; (2) Local blur detection: calculate the gradient magnitude decrease of the texture image after face reduction, and extract the edge intensity by Sobel or Laplacian operator. The larger the decrease, the more serious the blur; (3) Edge misalignment detection: extract the Canny edges of the texture images before and after face reduction, calculate the positional shift of the corresponding boundary points, and measure the degree of misalignment by pixel distance. When texture offset (recommended thresholds: low value [0,2] pixels, medium value [2,5] pixels, high value [5,15] pixels, and the recommended median value in this application is 3 pixels), local blur (recommended thresholds: gradient magnitude reduction value [0,15], medium value [15,30], high value [30,60], and the recommended median value in this application embodiment is 20 pixels), or edge misalignment (recommended thresholds: boundary offset low value [0,1] pixels, medium value [1,3] pixels, high value [3,8] pixels, and the recommended median value in this application is 2 pixels) exceeds the preset threshold, the first level of feedback is triggered to increase the protection weight and texture sampling accuracy of the area and reduce the subsequent facet reduction intensity; when the texture entropy is lower than the minimum retention threshold, the system assists in judging the potential risk of unrecoverable information in the area and triggers the second level of feedback—local rollback operation.
[0053] In another exemplary embodiment, the texture entropy described above describes the local information richness of a texture image, with a value range of [0,1].
[0054] Low values [0, 0.3]: The texture area lacks information and may have been over-compressed.
[0055] Median [0.3, 0.7]: Texture region information is well preserved.
[0056] High values [0.7, 1]: Rich texture details.
[0057] This invention recommends using a median value of 0.5 as the rollback trigger threshold for the following reasons: 0.5 is located in the middle of the range of texture entropy values, and is the dividing point from "insufficient" to "sufficient" texture information richness; When the local texture entropy is below 0.5, it indicates that more than half of the texture information in that area has been lost. Triggering a rollback can effectively avoid irreversible damage. Together with the normalized mean curvature threshold of 0.5 and the Sobel gradient magnitude threshold of 30, they form a unified parameter system. The best results are achieved when all three thresholds are set to the recommended values.
[0058] In addition to pixel distance, the texture offset detection mentioned above can also use structural similarity (SSIM) or peak signal-to-noise ratio (PSNR); in addition to gradient magnitude descent, local blur detection can also use Laplacian variance or frequency domain analysis; in addition to boundary offset, edge misalignment detection can also use Canny edge matching or contour coincidence.
[0059] In another exemplary embodiment, in step 105 above, when the total UV stretching amount is not greater than the UV protection threshold, the texture error is not greater than the texture error threshold, and the texture entropy is not greater than the texture entropy threshold, the detection result is determined to meet the surface reduction quality requirements; otherwise, the detection result is determined not to meet the surface reduction quality requirements.
[0060] In another exemplary embodiment, in existing face reduction methods, edge collapse is irreversible once performed, and cannot be recovered when local textures are severely lost. This application proposes a three-level progressive protection: the first level... The first level constrains texture errors; the second level adjusts in real time through closed-loop detection and differential feedback; the third level uses texture entropy to help determine potential unrecoverable information risks and triggers local rollback—undoing the most recent edge collapse in the region, restoring the original vertex positions, UV coordinates, and protection weights, while dynamically increasing the UV protection threshold of the region to prevent similar damage from recurring. These three progressive levels enhance the local recovery capability of the facet reduction process.
[0061] Step 105 above can be implemented using step 7 below.
[0062] Step 7: Irreversible damage avoidance mechanisms, including three levels of progressive protection: Level 1: Texture error pre-constraint. This is achieved through the texture mapping change term in the cost function. During the collapse decision-making stage, texture error is constrained to reduce the probability of texture damage.
[0063] Level 2: Closed-loop detection and differentiated feedback. When UV stretching exceeds the limit, the protection weight is increased and UV-related feedback is processed first. When texture error exceeds the limit, the sampling accuracy is increased and the edge collapse priority is reduced. The two detection methods employ different feedback strategies for different types of errors.
[0064] Level 3: Rollback due to irrecoverable information risk. When the local texture entropy falls below the minimum retention threshold, an auxiliary assessment is performed to determine the potential risk of irrecoverable information in that area, triggering a local rollback operation. The specific rollback execution order is as follows: (1) Vertex position recovery: Based on the backup data in the rollback record, restore the original three-dimensional coordinates of all vertices in the region; (2) UV coordinate restoration: restore the original UV texture coordinates corresponding to the vertices to ensure correct texture mapping; (3) Topology reconnection: restores the mesh topology relationship that has changed due to collapse to its original state to prevent topology errors such as holes and fragments; (4) Protection weight restoration: Restore the original protection weight value of the region to ensure that the region will not be selected again in subsequent face reduction. The specific implementation of the local rollback operation includes: recording the collapsed edges and their original vertex positions, UV coordinates and protection weights in each round of face reduction; when a region triggers an information irrecoverable risk warning, cancel the edges collapsed in the most recent round of the region, restore the original vertex positions, UV coordinates and protection weights, and mark the region as a prohibited collapse region; at the same time, multiply the UV protection threshold by an adjustment coefficient k greater than 1 (k>1) to dynamically increase the UV protection threshold of the region to prevent the same type of damage from happening again.
[0065] It is important to note that in existing polygon reduction methods, edge collapse is irreversible once executed, and severe local texture loss cannot be recovered. This application's embodiment uses texture entropy to assist in assessing potential irrecoverable information risks and triggering local rollback, improving the local recovery capability of the polygon reduction process. Through a three-level progressive irreversible damage avoidance mechanism, problems are intercepted step-by-step during polygon reduction: most texture errors are constrained at the first level, a small number are adjusted in real-time at the second level, and a very small number of irreversible damages are repaired at the third level. Topology consistency maintenance: During local rollback, the consistency of mesh topology relationships must be ensured. The specific implementation method is as follows: (1) Recording during collapse: When performing edge collapse, the topological relationships of the affected edges (including vertex connection relationships and edge-face dependency relationships) are recorded simultaneously. (2) Reconstruction during rollback: Based on the recorded topology, the original mesh topology is accurately reconstructed during rollback; (3) Error detection: After rollback, connectivity test is performed to ensure that there are no isolated vertices, hanging edges or fragmented surfaces; (4) Automatic repair: If a topology error is detected, the hole-filling algorithm is automatically invoked to repair it.
[0066] The aforementioned local rollback operation serves as an irreversible damage avoidance mechanism to address the problems in existing technologies where edge collapse is irreversible once executed and local texture loss is irrecoverable. This application performs UV stretching detection and texture baking quality detection simultaneously after each round of polygon reduction. These two processes work together: when UV stretching exceeds the limit, differential feedback adjusts the protection weight and polygon reduction priority; when the texture entropy falls below the rollback trigger threshold, the local rollback mechanism is triggered.
[0067] In addition to color information entropy, the texture entropy determination mentioned above can also use texture gradient entropy, frequency domain energy distribution, or structural similarity reduction to help determine the potential risk of irrecoverable information. In addition to canceling the most recent round of collapse, local rollback can also use multiple rounds of rollback or checkpoint snapshot recovery. In addition to multiplying by the adjustment coefficient k, the dynamic adjustment of the UV protection threshold can also use adaptive adjustment based on the texture entropy change rate or regional differential adjustment based on local features of the model.
[0068] In addition to threshold rules, adjustment strategies can also employ PID control, reinforcement learning strategies, or Bayesian optimization; in addition to fixed reduction, step size adjustment can also employ adaptive step size or linear search.
[0069] In another exemplary embodiment, steps 106 and 107 described above can be implemented using step 8.
[0070] Step 8: Iterative Output. Repeat steps 4 through 7 until preset output conditions are met, including achieving the target facet compression rate, meeting the threshold for structural fidelity, meeting the threshold for texture fidelity, or reaching the maximum number of iterations. When the conditions are met, output a lightweight, high-fidelity 3D simulation model.
[0071] In addition to the face number compression rate and fidelity threshold, the above-mentioned preset output conditions can also include total processing time limit, texture quality score lower limit, or convergence judgment of single-round error change.
[0072] In another exemplary embodiment, the process of the above-described non-standard object reduction optimization and texture baking collaborative processing method can be aided by an AI-assisted module. This AI-assisted module is an optional enhancement module of the technical solution of this application, rather than a necessary component of the core process. The core texture error closed-loop constraint and irreversible damage avoidance mechanism are entirely based on the real-time calculation of the current 3D simulation model and can run independently without relying on any AI module.
[0073] In another exemplary embodiment, the aforementioned AI-assisted module may include a dual-branch, three-channel feature extraction network (DTC-FEN) to assist in feature analysis. The first branch is a geometric feature branch, which takes a curvature distribution map and a normal vector field as input and outputs geometric key weights G. The second branch is a texture feature branch, which takes a texture color gradient map and a texture density map as input and outputs texture detail weights T. The third channel determines the display attention weights V based on the spatial constraint parameters of the 3D display scene. The AI-assisted module may also include a retrieval-enhanced parameter tuning (RAG) mechanism to recommend initial parameters, construct a historical case knowledge base, and retrieve the most similar cases to recommend parameters for the new model. When the AI-assisted module is not configured, all steps are completed in real-time through the current 3D simulation model and texture image. The role of AI assistance is to improve the accuracy of the analysis and the quality of the initial parameter recommendations, without changing the core workflow.
[0074] In addition to the DTC-FEN network, the AI-assisted module mentioned above can also employ other multi-branch feature extraction networks, Transformer architectures, or graph neural networks; the RAG mechanism, besides retrieval enhancement and parameter tuning based on case knowledge bases, can also employ collaborative filtering recommendation or meta-learning parameter initialization. The AI-assisted module is optional; when not configured, all steps are completed through real-time computation.
[0075] In another exemplary embodiment, the method described above can run independently without relying on a case database, and can record historical processing parameters and effects for recommending initial parameters for subsequent similar models. This module is only an optional auxiliary module and does not affect the independent operation of the core process.
[0076] Based on the same inventive concept, this application also provides a system for implementing the above-mentioned non-standard object surface reduction optimization and texture baking collaborative processing method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the system based on non-standard object surface reduction optimization and texture baking collaborative processing provided below can be found in the limitations of the non-standard object surface reduction optimization and texture baking collaborative processing method described above, and will not be repeated here.
[0077] In one exemplary embodiment, a collaborative processing system based on non-standard object surface reduction optimization and texture baking is provided, such as... Figure 4 As shown, the system includes: a 3D simulation model acquisition module, a feature analysis module, a protection weight generation module, a weighted edge collapse reduction surface module, a UV stretching judgment module, a texture baking quality detection module, an irreversible damage avoidance module, and a local error feedback adjustment module.
[0078] The 3D simulation model acquisition module is used to acquire high-precision 3D simulation models and corresponding texture images of non-standard objects to be processed, and to perform preprocessing, including denoising, hole filling, normal vector calculation, curvature calculation and scale normalization.
[0079] The feature analysis module is used to analyze the surface features of the model from three different technical dimensions simultaneously, including the geometric key region analysis submodule, the texture detail region analysis submodule, and the display attention region analysis submodule based on the spatial constraints of the 3D display scene, providing a basis for differentiated surface reduction.
[0080] The protection weight generation module is used to synthesize the analysis results of the three dimensions into a unified "protection map" and generate a multi-layer protection weight map W=αG+βT+γV, so as to realize differentiated protection of different areas according to their importance.
[0081] The weighted edge collapse reduction surface module is used to construct a weighted edge collapse cost function based on the protection weight graph. Perform face reduction, texture mapping changes By incorporating texture error into the collapse decision, the three costs work together to determine the collapse order.
[0082] The UV stretching judgment module detects the degree of texture mapping deformation after each round of surface reduction. When it exceeds the UV protection threshold, it marks an anomaly and triggers feedback, which is the first detection in the closed loop.
[0083] The texture baking quality detection module is used to detect texture offset, blur, and edge misalignment. At the same time, it calculates the local texture entropy. When the texture error exceeds the standard, it triggers the first level of feedback. When the texture entropy is lower than the minimum retention threshold, it triggers the risk warning of unrecoverable information. It is the second detection in the closed loop.
[0084] The irreversible damage avoidance module is used to implement three levels of progressive protection—Level 1 through The module employs a pre-constraint texture error mechanism, a second level of real-time adjustment through closed-loop detection and differential feedback, and a third level of triggering local rollback and dynamically adjusting the UV protection threshold based on texture entropy determination. This module includes a collapse recording submodule, a rollback execution submodule, and a threshold dynamic adjustment submodule.
[0085] The local error feedback adjustment module is used to feed back the differential results of UV stretching judgment and texture quality detection to the reduction strategy. Different adjustment strategies are adopted for different types of errors, and the reduction step size, protection weight and texture sampling accuracy are comprehensively adjusted to form a closed loop.
[0086] In another exemplary embodiment, the system also includes an AI-assisted module.
[0087] AI-assisted module: It uses the DTC-FEN network to assist in feature analysis and the RAG mechanism to recommend initial parameters. It is an optional enhancement module. When not configured, it completes all steps through real-time calculation without affecting the independent operation of the core process.
[0088] In another exemplary embodiment, the system further includes a case parameter recording module.
[0089] Case parameter recording module: Records historical processing parameters and effects, used to recommend initial parameters for subsequent similar models. It is an optional auxiliary module and does not affect the core process.
[0090] Existing methods for reducing the number of faces in 3D simulation models of non-standard objects only include a geometric error term in their cost function, and texture errors do not affect the collapse decision; face reduction and texture baking are processed in separate steps; edge collapse is irreversible once performed. The core difference between this application and existing technologies is that it incorporates texture mapping variation terms. By incorporating a collapse cost function to pre-constrain texture errors, and employing a three-level progressive irreversible damage avoidance mechanism, partial recovery of texture damage is achieved. This application addresses several issues in existing non-standard object 3D simulation models, including: texture errors not affecting collapse decisions leading to uncontrollable texture damage; irreversible edge collapse resulting in difficulty in recovering texture damage; separate processing of facet reduction and texture baking leading to difficulty in real-time texture distortion repair; and the lack of UV stretching closed-loop feedback. Compared to existing technologies, this application has the following significant advantages: 1. Achieving pre-emptive protection of texture fidelity through texture error closed-loop constraint edge collapse. Existing methods first reduce surfaces and then detect texture problems, with texture errors not affecting the collapse decision. This application directly incorporates the texture mapping change term Ct(e) into the collapse cost function, allowing texture errors to play a role in the collapse decision stage—edges with large texture changes have higher collapse costs and are ranked later, thus reducing the probability of texture damage. Results: Significantly improved texture fidelity, and a substantial reduction in texture offset and edge misalignment problems.
[0091] 2. A three-tiered progressive irreversible damage avoidance mechanism enables partial recovery of texture damage during the facet reduction process. In existing facet reduction methods, edge collapse is irreversible once performed, and severe local texture loss cannot be recovered. This application proposes a three-tiered progressive protection mechanism: the first tier uses Ct(e) to pre-constrain texture errors; the second tier uses closed-loop detection and differential feedback for real-time adjustment; and the third tier uses texture entropy determination to trigger local rollback—undoing the most recent edge collapse in that region and restoring the original data, while dynamically increasing the UV protection threshold. Effect: The facet reduction process becomes "recoverable," significantly reducing the probability of irreversible texture damage and providing local recovery capabilities.
[0092] 3. Differentiated surface reduction through 3D fusion weighted protection. The protection requirements in three dimensions originate from three different technical fields: geometric structure, image processing, and human-computer interaction. Features are extracted from each technical dimension to improve the completeness of the protection weight representation. The three dimensions are fused to generate a multi-layered protection weight map, which is then incorporated into the collapse cost function, resulting in less surface reduction in important areas and more surface reduction in less important areas. Results: Significantly improved retention of key details such as local protrusions, hole edges, natural textures, and highlighted areas.
[0093] 4. Real-time feedback is achieved by judging individual ring formation through UV stretching. Existing methods only discover UV stretching problems after the surface reduction is completed, resulting in high repair costs. This application judges the UV stretching amount separately after each round of surface reduction, and immediately provides feedback and adjustment when problems are found, preventing the accumulation of problems and enabling the detection and repair of UV stretching anomalies during the surface reduction process.
[0094] 5. It can run independently without relying on the AI module. The core closed-loop constraints and irreversible damage avoidance mechanism are calculated in real time based entirely on the current 3D simulation model. The AI-assisted module is an optional enhancement, which lowers the deployment threshold and allows it to run independently without an AI model.
[0095] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0096] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for non-uniform object area reduction optimization and texture baking co-processing, characterized in that, include: The original 3D simulation model of the non-standard object is used as the 3D simulation model before the first round of face reduction, and the value of the face reduction round n is set to 1. The collapse cost of different local regions of the 3D simulation model before the nth round of reduction is calculated using the weighted edge collapse cost function. Based on the collapse cost of different local regions, the 3D simulation model before the nth round of reduction is reduced in terms of surface area to obtain the 3D simulation model after the nth round of reduction in surface area. UV stretching detection and texture baking quality detection are performed on the 3D simulation model after the nth round of surface reduction to obtain the detection results of different local areas; When the detection results of one or more local regions do not meet the surface reduction quality requirements, a parameter adjustment operation or a local rollback operation is performed. The 3D simulation model before the nth round of surface reduction or the 3D simulation model after the local rollback operation is used as the new 3D simulation model before the nth round of surface reduction, and the process returns to the step of "calculating the collapse cost of different local regions of the 3D simulation model before the nth round of surface reduction using the weighted edge collapse cost function". When the detection results of each local region meet the surface reduction quality requirements but do not meet the preset output conditions, the 3D simulation model after surface reduction in the nth round is used as the 3D simulation model before surface reduction in the (n+1)th round. The value of n is increased by 1, and the process returns to the step of "calculating the collapse cost of different local regions of the 3D simulation model before surface reduction in the nth round using the weighted edge collapse cost function". When the detection results of each local area meet the surface reduction quality requirements and reach the preset output conditions, the final surface reduction 3D simulation model is output as the lightweight 3D simulation model.
2. The non-uniform object area-reduction optimization and texture baking co-processing method of claim 1, wherein, The weighted edge collapse cost function is: ; wherein, is a collapse cost for the local region, is a geometry error term for the local region, is a protection weight penalty term for the local region, , is a geometry key weight for the local region, is a texture detail weight for the local region, is a display focus weight for the local region, , , is a weight coefficient, is a texture mapping variation term for the local region, , is an adjustment coefficient. 3. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 2, characterized in that, The geometric key weights, texture detail weights, and display attention weights of each local region are obtained by performing three-dimensional feature analysis on the original 3D simulation model; the three-dimensional features include vertex curvature, texture image color gradient, and spatial observation constraint parameters.
4. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 1, characterized in that, The detection results include: comprehensive UV stretching amount, texture error, and texture entropy; When the total UV stretching amount is not greater than the UV protection threshold, the texture error is not greater than the texture error threshold, and the texture entropy is not greater than the texture entropy threshold, the detection result is determined to meet the surface reduction quality requirements; otherwise, the detection result is determined not to meet the surface reduction quality requirements.
5. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 1, characterized in that, The formula for calculating the overall UV stretching amount is: ; ; ; ; in, The total UV stretching amount for local region e. , , These represent the UV area stretching, UV side length stretching, and UV angular distortion rate of the local region e, respectively. , To reduce the UV area of the local region e before and after the front, , To reduce the UV edge length of the local region e before and after the front, , To reduce the UV angle of the local region e before and after the front, , , These are the weighting coefficients.
6. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 4, characterized in that, The local area where the detection result does not meet the preset conditions is taken as the target local area. For each target local area, the parameter adjustment operation or local rollback operation specifically includes: When the total UV stretching of the target local area is greater than the UV protection threshold, and the texture error is not greater than the texture error threshold, and the texture entropy is not greater than the texture entropy threshold, then the first parameter adjustment operation is performed on the target local area. If the texture error of the target local region is greater than the texture error threshold and the texture entropy is not greater than the texture entropy threshold, then the second parameter adjustment operation is performed on the target local region. If the texture entropy of the target local region is greater than the texture entropy threshold, a local rollback operation is performed on the target local region.
7. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 6, characterized in that, The first parameter adjustment operation includes: Increase the protection weight penalty for the target local area; Or reduce the step size of the reduction area; Alternatively, UV remapping can be performed on a local area of the target.
8. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 6, characterized in that, The second parameter adjustment operation includes: Increase the protection weight penalty term for the target local area and decrease the reduction step size.
9. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 6, characterized in that, The partial rollback operation includes: In the nth round of reduction, the vertex positions, UV coordinates, mesh topology, and protection weight penalty terms of the target local region in the 3D simulation model are adjusted sequentially to restore the vertex positions, UV coordinates, mesh topology, and protection weight penalty terms of the target local region to their original state, and the UV protection threshold of the target local region is increased.
10. The method for co-processing non-standard object surface reduction optimization and texture baking according to claim 6, characterized in that, The preset output conditions include: The face reduction ratio reaches the target value or the number of face reduction cycles reaches the upper limit.
