AI analogue simulation method and system for facial beauty and plastic surgery

Abstract

The invention discloses an AI simulation method and system for facial cosmetic plastic surgery, and the method comprises the steps: for a specific region with high feature complexity, generating a feature subset containing fine feature distribution through high-performance node distribution processing, combination with an adaptive feature extraction algorithm, and dynamic adjustment of extraction frequency and segmentation granularity; on the basis, a three-dimensional reconstruction algorithm is adopted to dynamically adjust the splicing weight, particularly, the splicing precision is improved and a high-precision three-dimensional model is generated for an area with relatively thin tissue thickness, and finally, a vivid effect simulation diagram is obtained through rendering parameter adjustment driven by a user demand and preferentially rendering a height modification area. According to the method, through adaptive parameter adjustment and fine processing, the modeling precision and rendering efficiency of the complex biological characteristic data are remarkably improved, and the method is suitable for medical images, biological recognition and other scenes.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular discloses an AI simulation method and system for facial cosmetic surgery. Background Technology

[0002] The field of facial cosmetic surgery has garnered significant attention in recent years due to a surge in personalized demands, particularly with the integration of internet technology and artificial intelligence. Simulating post-operative results has become crucial for enhancing user experience and decision-making quality. Predicting surgical outcomes digitally not only helps users intuitively understand potential changes but also provides medical institutions with precise reference data. Technological advancements in this field directly impact user satisfaction and industry efficiency; therefore, developing efficient, accurate, and safe simulation systems is of paramount importance. However, existing technologies still face significant challenges in balancing meeting diverse user needs with ensuring data security, requiring breakthroughs to drive industry development.

[0003] Current facial plastic surgery simulation methods largely rely on traditional image processing or simple 3D modeling, which struggles to adequately address complex facial features and personalized needs. Many approaches are limited to a single data source during the data acquisition phase, failing to comprehensively capture multidimensional facial information, such as subtle differences in skin texture, bone structure, and soft tissue distribution. This results in simulations that often lack realism and fail to meet users' precise expectations for postoperative outcomes.

[0004] The complexity of facial features requires systems capable of adjusting scanning parameters based on the characteristics of different regions. For example, the eye area requires high-frequency scanning to capture subtle textures, while the cheekbone area requires low-frequency scanning to highlight bone contours. Existing technologies often employ fixed scanning strategies, making it difficult to adapt to these differences, resulting in models that lack realistic detail. For instance, when simulating the effects of eyelid surgery, the system may fail to accurately distinguish the subtle differences between the eyelids and the corners of the eyes, generating simulations that do not meet user expectations. This inaccurate modeling makes users hesitant to trust the simulation results, impacting their decision-making confidence.

[0005] In actual business scenarios, for example, when users want to preview the effects of rhinoplasty through simulation, if the system cannot dynamically allocate processing resources according to the differences between the soft tissue and bone structure of the nose, the generated model may be distorted at the alar or bridge of the nose, affecting the user's judgment of the postoperative effect.

[0006] Therefore, how to achieve efficient dynamic data allocation and accurate 3D modeling has become a key problem that this research urgently needs to solve. Summary of the Invention

[0007] This invention provides an AI simulation method and system for facial cosmetic surgery, aiming to solve at least one of the defects existing in the prior art.

[0008] One aspect of the present invention relates to an AI simulation method for facial cosmetic surgery, comprising the following steps: S100. Acquire multi-source biometric data from the data acquisition device, classify the feature complexity of the multi-source biometric data, and if the multi-source biometric data contains fine details of a specific region, allocate high-performance nodes for processing to obtain a classified multidimensional dataset, wherein the multi-source biometric data contains image information, depth information and thermal imaging details. S200. Based on the classified multidimensional dataset, a feature extraction algorithm is used to extract key feature points. The parameters are adaptively adjusted according to the complexity of different regions. If the feature distribution in a specific region is dense, the extraction frequency is increased to obtain an initial feature map containing the feature distribution. S300. Obtain segmentation information from the initial feature map, use a segmentation algorithm to divide different regions, adjust the segmentation granularity according to structural differences, and refine the grid if the boundary of the segmented region changes significantly to obtain the feature subset of the sub-region. S400. Based on the feature subsets of the sub-regions, a three-dimensional reconstruction algorithm is used to generate a model. The stitching weights are dynamically adjusted for the scanning parameters of each region. If the tissue thickness of a specific region is less than the threshold, the stitching accuracy is improved to obtain a high-precision three-dimensional model. S500: Obtain rendering data from a high-precision 3D model, combine it with preset effect parameters to perform simulation rendering, adjust lighting and texture according to user needs, and if the user needs include height modification, prioritize rendering the relevant areas to obtain the final effect simulation image.

[0009] Further, step S100 includes: S110. Receive multi-source biometric data containing image information, depth information and thermal imaging details from the data acquisition device, and standardize the multi-source biometric data through a preset sensor interface to obtain a multi-source data set in a unified format. S120. For multi-source datasets, feature extraction algorithms are used to quantitatively analyze the feature complexity of image information, depth information and thermal imaging details. If the feature complexity exceeds a preset threshold, it is determined to be a high-complexity feature, and the feature complexity level after classification is obtained. S130. Based on the feature complexity level, if it contains fine details of a specific region, the high-complexity features are distributed to high-performance nodes for processing through a load balancing algorithm to obtain an optimized multidimensional feature dataset. S140. A data fusion algorithm is used to integrate the image information, depth information and thermal imaging details in the multidimensional feature dataset to generate a multidimensional dataset.

[0010] Further, step S200 includes: S210. Obtain multidimensional data containing image information, depth information and thermal imaging details from the classified multidimensional dataset. Process the multidimensional data using a preset feature extraction algorithm to extract key feature points and obtain the first feature point set. S220. For the first set of feature points, a quantitative evaluation method is used to analyze the feature point density of different regions. If the feature point density of a region exceeds a preset threshold, the region is determined to be a high-complexity region, and the region complexity distribution is obtained. S230. Based on the distribution of regional complexity, the feature extraction frequency of high-complexity regions is dynamically increased through an adaptive parameter adjustment algorithm, while the default frequency is maintained for low-complexity regions, to obtain an optimized set of second feature points. S240. Use a feature mapping algorithm to spatially integrate the key feature points in the second feature point set, map image information, depth information and thermal imaging details to a unified coordinate system, and generate an initial feature map containing feature distribution.

[0011] Further, step S300 includes: S310. Obtain segmentation information from the initial feature map, and use a region-growing-based segmentation algorithm to divide the initial feature map into regions to obtain the first region set; S320. For the first set of regions, the degree of boundary change of each region is calculated by gradient analysis algorithm. If the degree of boundary change exceeds the preset threshold, it is determined as a high-change region, and the boundary change distribution is obtained. S330. Based on the boundary change distribution, an adaptive grid partitioning algorithm is used to increase the grid density in high-change areas and maintain the default grid density in low-change areas to obtain the second region set. The second region set is obtained using the following formula:

[0012] in, Represents the second region set. and Indicates the first The coordinates of each grid node, Represents the set of natural numbers. Indicates the first Mesh density at each node Indicates the grid density threshold; S340. Extract feature points from the clusters in the second region set, and use a feature clustering algorithm to perform regional clustering on the extracted feature points in the clusters to obtain the final feature subset set.

[0013] Further, step S400 includes: S410. Obtain the regional feature distribution from the feature subset, and use the stereomicroscope scanning algorithm to collect three-dimensional data of each region. If the tissue thickness of the region is lower than the preset threshold, increase the scanning frequency to obtain a high-resolution three-dimensional data point set. S420. Based on the high-resolution three-dimensional data point set, an initial three-dimensional mesh is generated using a voxelization algorithm. If the boundary variation of the mesh cells in the initial three-dimensional mesh exceeds a preset threshold, the local mesh of the region is refined to obtain an optimized three-dimensional mesh. S430. Extract surface feature points from the optimized 3D mesh, and use the weighted least squares method to stitch together the extracted surface feature points and assign weights. If the tissue thickness of the region is lower than the preset threshold, increase the weight assignment ratio to obtain a weighted feature point set. S440. Based on the weighted feature point set, a preliminary three-dimensional model is generated using a volume rendering algorithm. If the surface curvature of the preliminary three-dimensional model exceeds a preset threshold, the surface curvature exceeding the preset threshold is locally smoothed to obtain a high-precision three-dimensional model.

[0014] Further, step S500 includes: S510. Obtain regional geometric data from a high-precision 3D model, and use a stereomicroscope scanning algorithm to extract local feature distribution. If the curvature of the local features exceeds a preset threshold, increase the sampling frequency to obtain a high-resolution geometric dataset. S520. Based on the high-resolution geometric dataset, the initial lighting and shadow data is generated using a stereo rendering algorithm. The rendering brightness is adjusted in combination with preset effect parameters to obtain a preliminary lighting and shadow rendering image. S530. Extract surface texture points from the initial lighting and shadow rendering map. If the user's requirements include height modification, use the weighted least squares method to assign priority weights to the texture points to obtain a weighted texture point set. S540. Based on the weighted texture point set, a volume rendering algorithm is used to generate a simulation image. If the smoothness of the simulation image is lower than a preset threshold, local smoothing optimization is performed on the relevant areas to obtain the final rendering image.

[0015] Another aspect of the present invention relates to an AI simulation system for facial cosmetic surgery, used to execute the aforementioned AI simulation method for facial cosmetic surgery, comprising: The multidimensional dataset acquisition module is used to acquire multi-source biometric data from the data acquisition device, classify the feature complexity of the multi-source biometric data, and if the multi-source biometric data contains fine details of a specific region, it is assigned a high-performance node for processing to obtain the classified multidimensional dataset, in which the multi-source biometric data contains image information, depth information and thermal imaging details. The initial feature map acquisition module is used to extract key feature points from the classified multidimensional dataset using a feature extraction algorithm. It adaptively adjusts parameters according to the complexity of different regions. If the feature distribution in a specific region is dense, the extraction frequency is increased to obtain an initial feature map containing the feature distribution. The feature subset acquisition module is used to obtain segmentation information from the initial feature map, divide different regions using a segmentation algorithm, adjust the segmentation granularity according to structural differences, and refine the grid if the boundary of the segmented region changes significantly to obtain the feature subset of the region. The high-precision 3D model acquisition module is used to generate a model based on the feature subsets of the sub-regions using a 3D reconstruction algorithm. It dynamically adjusts the stitching weights of the scanning parameters of each region. If the tissue thickness of a specific region is less than a threshold, the stitching accuracy is improved to obtain a high-precision 3D model. The effect simulation image acquisition module is used to obtain rendering data from high-precision 3D models, combine it with preset effect parameters to perform simulation rendering, adjust lighting and texture according to user needs, and if the user needs include height modification, the relevant areas will be rendered first to obtain the final effect simulation image.

[0016] The beneficial effects achieved by this invention are as follows: This invention provides an AI simulation method and system for facial cosmetic surgery. Addressing the complexity differences in image information, depth information, and thermal imaging details within multi-source biometric data, it proposes a solution that dynamically adjusts the processing strategy. For specific regions with high feature complexity, high-performance nodes are allocated for processing, combined with an adaptive feature extraction algorithm to dynamically adjust the extraction frequency and segmentation granularity, generating a feature subset containing finely distributed features. Based on this, a 3D reconstruction algorithm is used to dynamically adjust the stitching weights, particularly improving stitching accuracy in thinner tissue regions, generating a high-precision 3D model. Finally, user-driven rendering parameter adjustments prioritize rendering regions with height modifications, resulting in a realistic simulation image. This invention significantly improves the modeling accuracy and rendering efficiency of complex biometric data through adaptive parameter adjustment and refined processing, making it suitable for scenarios such as medical imaging and biometric recognition. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of an AI simulation method for facial cosmetic surgery according to the present invention. Figure 2 This is a functional block diagram of an embodiment of an AI simulation system for facial cosmetic surgery according to the present invention.

[0018] Explanation of icon numbers: 10. Multidimensional dataset acquisition module; 20. Initial feature map acquisition module; 30. Feature subset acquisition module; 40. High-precision 3D model acquisition module; 50. Effect simulation image acquisition module. Detailed Implementation

[0019] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0020] like Figure 1 As shown, the first embodiment of the present invention proposes an AI simulation method for facial cosmetic surgery, including the following steps: Step S100: Obtain multi-source biometric data from the data acquisition device, classify the feature complexity of the multi-source biometric data, and if the multi-source biometric data contains fine details of a specific region, assign high-performance nodes for processing to obtain a classified multidimensional dataset, wherein the multi-source biometric data contains image information, depth information and thermal imaging details.

[0021] Multi-source biometric data refers to a collection of multi-dimensional data acquired simultaneously by multi-modal data acquisition devices (such as RGB cameras, depth sensors, and infrared thermal imagers) to comprehensively depict the physiological morphology and tissue characteristics of the face. It primarily comprises three modalities: image information (2D surface visual features), depth information (3D structural features), and thermal imaging details (infrared thermal radiation tissue features). Its core value lies in overcoming the limitations of single-modal data (such as the lack of stereoscopic information in 2D images and the absence of texture details in depth data), providing raw data support across the three dimensions of "surface-stereoscopic-tissue" for facial cosmetic surgery AI simulation, ensuring the realism and accuracy of subsequent simulations.

[0022] Based on this, by classifying features by complexity (distinguishing between conventional features and fine detail features), data containing fine details in specific regions (requiring high computing power for parsing) are allocated to high-performance nodes for processing. The resulting structured dataset is the classified multidimensional dataset. Essentially, it is a multimodal data system layered according to "feature complexity + processing requirements," which ensures efficient processing of conventional features while parsing fine details through high-performance nodes, laying the data foundation for high-precision simulation of facial cosmetic surgery.

[0023] Step S200: Based on the classified multidimensional dataset, use a feature extraction algorithm to extract key feature points, adaptively adjust parameters according to the complexity of different regions, and increase the extraction frequency if the feature distribution in a specific region is dense, to obtain an initial feature map containing the feature distribution.

[0024] An initial feature map containing feature distribution refers to a multi-dimensional dataset (containing image information, depth information, and a subset of regular / fine features of thermal imaging details) as input. It extracts key facial feature points (covering surface texture, three-dimensional morphology, and tissue thermal properties) using feature extraction algorithms adapted to multi-modal data (such as SIFT, 3D Harris, and thermal gradient extremum detection). Extraction parameters are dynamically adjusted according to the feature complexity of different regions (increasing extraction frequency and narrowing the extraction window in densely distributed areas, and vice versa in sparse areas). The final result is a multi-modal feature map that characterizes the location, type, density, and relationships of feature points in the form of a "spatial grid + feature attribute matrix." This initial feature map transforms the "key features" in multi-dimensional data into "structured, localizable, and quantifiable" distribution information. It preserves the feature differences between different facial regions (such as fine texture around the eyes vs. smooth areas of the cheeks) and provides "feature density guidance" and "initial localization benchmarks" for subsequent region segmentation and high-precision 3D reconstruction, meeting the dual simulation requirements of facial cosmetic surgery for "detail accuracy" and "overall morphology."

[0025] Step S300: Obtain segmentation information from the initial feature map, use a segmentation algorithm to divide different regions, adjust the segmentation granularity according to structural differences, and refine the grid if the boundary of the segmented region changes significantly to obtain the feature subset of the sub-region.

[0026] Feature subsets for different regions refer to extracting segmentation reference information (such as feature point density, boundary contours, and structural attributes) from an initial feature map containing feature distributions. The face is then divided into sub-regions using segmentation algorithms adapted to facial anatomy (such as medically optimized U-Net and Mask R-CNN). Simultaneously, the segmentation granularity is dynamically adjusted based on the structural differences between different regions (e.g., distinguishing between skeletal and soft tissue areas, and between smooth and complex boundaries). If significant morphological changes are detected at the boundaries of segmented regions (e.g., the three-dimensional transition between the bridge and root of the nose, or the concave transition between the eye socket and eyelid), the segmentation mesh is refined to preserve boundary details. The final result is a structured set of features categorized as "independent sub-regions." Feature subsets for different regions decompose the global features of the initial feature map into local feature units that are "functionally related and structurally similar" (e.g., eye subsets, nose subsets), providing precise local feature support for subsequent regional 3D reconstruction and personalized plastic surgery modifications.

[0027] Step S400: Based on the feature subsets of the sub-regions, a 3D reconstruction algorithm is used to generate a model. The stitching weights of the scanning parameters of each region are dynamically adjusted. If the tissue thickness of a specific region is less than the threshold, the stitching accuracy is improved to obtain a high-precision 3D model.

[0028] A high-precision 3D model refers to a digital model that takes a subset of features from different regions (including image texture, depth, morphology, and thermal imaging tissue features of sub-regions such as the eyes, nose, and lips) as input, and dynamically adjusts the stitching weights of the scanning parameters (point cloud density, registration accuracy) of each sub-region through 3D reconstruction algorithms adapted to different facial anatomical structures (such as Poisson reconstruction in the skeletal region and dynamic surface reconstruction in the soft tissue region). If a specific region is detected as thin tissue (such as the eyelid or nasal tip, with a thickness ≤2mm), the point cloud registration accuracy and stitching weight of that region are increased. The final generated model can restore the 3D morphology of the face's "skeletal framework + soft tissue layers + surface texture" with local detail errors ≤0.1mm. High-precision 3D models solve the problem of reconstruction distortion in thin facial tissue areas through "regional reconstruction + dynamic weight stitching," accurately depicting the three-dimensional morphology and tissue layers of the face, and providing a high-fidelity digital twin carrier for "preoperative planning, effect simulation, and surgical navigation" in facial cosmetic surgery.

[0029] Step S500: Obtain rendering data from the high-precision 3D model, perform simulation rendering in combination with preset effect parameters, adjust lighting and texture according to user requirements, and if the user requirements include height modification, prioritize rendering the relevant areas to obtain the final effect simulation image.

[0030] The final effect simulation image refers to a rendering data system based on a high-precision 3D model. It extracts surface morphology (such as nasal bridge curvature and eyelid folds), tissue layers (such as skin / fat layer thickness), and texture coordinates (such as pores and skin tone distribution) from the model. This data is then combined with preset effect parameters for facial cosmetic surgery (such as "natural rhinoplasty height," "natural double eyelid width," and "skin tone brightening gradient") for basic rendering. The rendering strategy is then dynamically adapted based on the user's personalized needs (such as height modification, texture optimization, and lighting style adjustment). If the needs include height / shape modifications to the nasal bridge or jaw angle, priority is given to fine-tuning these areas (such as adjusting lighting angles to highlight shape changes and optimizing texture fit). The final result is a multi-view (front / side / 45° angle), highly accurate (≥95% similarity to the actual post-operative effect) visual cosmetic effect map. The final effect simulation image transforms "abstract cosmetic surgery needs" into "concrete visual effects," preserving the natural transition of facial features while accurately presenting the user's customized cosmetic surgery results, providing an intuitive basis for pre-operative planning communication and user decision-making.

[0031] Furthermore, the AI ​​simulation method for facial cosmetic surgery proposed in this embodiment includes step S100: Step S110: Receive multi-source biometric data containing image information, depth information, and thermal imaging details from the data acquisition device, and standardize the multi-source biometric data through a preset sensor interface to obtain a multi-source data set in a unified format.

[0032] A unified format for a multi-source dataset is derived using the following formula: (1) In formula (1), This represents a multi-source data set in a unified format. Indexes representing three types of biometric data: image information, depth information, and thermal imaging details. Indicates the first Weighting coefficients for different data types Indicates the first Standardized processing functions for sensor interfaces corresponding to various data types. Indicates the first A variety of original biometric data sources, among which When the value is 1, it corresponds to the image information. When the value is 2, it corresponds to depth information. When the value is 3, it corresponds to thermal imaging details.

[0033] In facial recognition scenarios, data acquisition devices include visible light cameras, depth sensors, and thermal imagers, which respectively collect facial images, 3D depth information, and thermal imaging data.

[0034] Standardization is achieved through a preset sensor interface. Specifically, visible light images are uniformly adjusted to 1080p resolution, depth information is converted to a unified depth map format, and thermal imaging data is normalized to a grayscale value range of 0-255.

[0035] Such standardization ensures consistent data formats across multiple sources, facilitating subsequent processing and significantly improving data compatibility and algorithm efficiency.

[0036] Feature extraction algorithms quantify the feature complexity of image information, depth information, and thermal imaging details. For facial images, texture features, such as the fineness of skin texture, are extracted; depth information is used to extract facial contour curvature; and thermal imaging is used to extract temperature distribution features. Assuming a preset complexity threshold of 1000, after complexity quantification, if the texture feature value of a certain region reaches 1200, it is classified as a high-complexity feature. This classification method effectively filters features that require priority processing, optimizes the allocation of computational resources, and improves processing accuracy.

[0037] Step S120: For the multi-source dataset, a feature extraction algorithm is used to quantitatively analyze the feature complexity of image information, depth information and thermal imaging details. If the feature complexity exceeds a preset threshold, it is determined to be a high-complexity feature, and the feature complexity level after classification is obtained.

[0038] The feature complexity of image information is derived by the following formula: (2) In formula (2), The quantization value representing the feature complexity of image information. and These represent the height and width of the image in pixels, respectively. Indicates position Pixel value at that location, and These respectively indicate that the position is in direction and The gradient value in the direction is used to quantify the complexity of image features by calculating the average gradient magnitude across the entire image. The feature complexity of depth information is derived from the following formula: (3) In formula (3), The quantization value representing the feature complexity of depth information. This represents the total number of regions in the depth image. Indicates the first The variance of depth values ​​within each region block Indicates the first The mean depth value within each region block This represents a small constant to prevent division by zero, and the complexity of the depth feature is evaluated by calculating the coefficient of variation of the depth change of each region block; Set the criteria for determining the feature complexity level after classification: (4) In formula (4), This indicates the level of feature complexity after classification. This represents the numerical value of feature complexity calculated using the feature extraction algorithm. This represents a preset complexity threshold. When the feature complexity exceeds the threshold, it is judged as a high-complexity feature; otherwise, it is a low-complexity feature, thus achieving the classification of complexity levels of features from multiple sources.

[0039] Step S130: Based on the feature complexity level, if it contains fine details of a specific region, the high-complexity features are distributed to high-performance nodes for processing using a load balancing algorithm to obtain an optimized multidimensional feature dataset.

[0040] The quality assessment value of the optimized multidimensional feature dataset is obtained using the following formula:

[0041] In formula (5), This represents the quality assessment value of the optimized multidimensional feature dataset. This represents the total number of features after processing. Indicates the first The importance factors of each feature Indicates the first Information entropy of each feature Indicates the first The processing quality coefficient of each feature.

[0042] High-complexity features are distributed to high-performance nodes for processing using a load balancing algorithm. For example, in a face recognition system, the complex features of the eye region, which contains fine details, need to be allocated to a GPU cluster for parallel processing, while low-complexity regions such as the background are processed by ordinary nodes. The load balancing algorithm dynamically allocates computing tasks based on task priority and node load, ensuring that the overall system efficiency is improved by more than 30% while reducing processing latency.

[0043] Step S140: Use a data fusion algorithm to integrate the image information, depth information and thermal imaging details in the multidimensional feature dataset to generate a multidimensional dataset.

[0044] Multidimensional data sets are derived using the following formula: (6) In formula (6), This represents the integrated cube. A numerical vector representing image information. A numerical vector representing depth information. Numerical vectors representing details in thermal imaging. The standard deviation representing image information The standard deviation representing depth information The standard deviation of thermal imaging information.

[0045] Data fusion algorithms integrate multidimensional feature datasets. For face recognition, a weighted fusion method is used to integrate image information, depth information, and thermal imaging details with weights of 0.4, 0.3, and 0.3 to generate a multidimensional dataset. For example, the fused dataset comprehensively reflects the appearance, three-dimensional structure, and temperature distribution features of a face, improving accuracy to over 98% when used for identity verification. The fusion process considers the complementarity of multi-source data, significantly enhancing the robustness of features and reducing the impact of ambient light or occlusion on recognition.

[0046] Preferably, the AI ​​simulation method for facial cosmetic surgery proposed in this embodiment includes step S200: Step S210: Obtain multidimensional data containing image information, depth information and thermal imaging details from the classified multidimensional dataset. Process the multidimensional data using a preset feature extraction algorithm to extract key feature points and obtain the first feature point set.

[0047] The first set of feature points is obtained by the following formula: (7) In formula (7), Indicates the first The fused multidimensional feature values, Indicates position Image information intensity at that location Indicates position Depth information value at that location, Indicates position Thermal imaging detail data at the location, , , These represent the fusion weighting coefficients for the three data types: image information, depth information, and thermal imaging details.

[0048] In face recognition scenarios, extracting data containing image information, depth information, and thermal imaging details from the classified multidimensional dataset requires ensuring the accuracy and efficiency of feature extraction. For the feature extraction algorithm, a convolutional neural network combined with keypoint detection technology is employed to extract texture features from face images, contour features from depth information, and temperature distribution features from thermal imaging. Specifically, for 1080p face images, the algorithm identifies the pixel distribution of key areas such as the eyes and nose; depth information is extracted from point cloud data to obtain facial curvature key points; and thermal imaging data identifies temperature anomalies, such as marking points where the temperature around the eyes is 5 degrees Celsius above the average as key points. These key points form the first feature point set, containing approximately 5000 feature points, covering the main facial areas.

[0049] Step S220: For the first set of feature points, a quantitative evaluation method is used to analyze the feature point density of different regions. If the feature point density of a region exceeds a preset threshold, the region is determined to be a high-complexity region, and the region complexity distribution is obtained.

[0050] The density of feature points in a region is obtained using the following formula: (8) In formula (8), Indicates the first Feature point density of each region Indicates the first The number of feature points in a region Indicates the first The size of each region.

[0051] Define the conditions used to determine the complexity category of a region: (9) In formula (9), Indicates the first Complexity identifier for each region This represents the preset density threshold, when the first... Feature point density of each region Exceeding the preset density threshold The time marker is designated as a high-complexity region.

[0052] When quantifying feature point density, a threshold of 100 feature points per square centimeter is set. For example, the feature point density in the eye region reaches 150; if the feature point density exceeds the threshold, it is classified as a high-complexity region. The cheek region, however, has a density of only 50 feature points, classifying it as a low-complexity region. The distribution of region complexity is visualized using a density heatmap, intuitively reflecting the feature point distribution. This quantification method facilitates subsequent dynamic adjustments to processing strategies and optimization of resource allocation.

[0053] Step S230: Based on the distribution of regional complexity, the feature extraction frequency of high-complexity regions is dynamically increased through an adaptive parameter adjustment algorithm, while the default frequency is maintained for low-complexity regions, to obtain the optimized second feature point set.

[0054] The feature extraction frequency for high-complexity regions is dynamically increased using the following formula: (10) In formula (10), Indicates the first Feature extraction frequency of each region This indicates the default base frequency. Indicates the frequency adjustment factor. Indicates the exponential growth parameter. Indicates the first Complexity assessment value for each region, This represents the minimum complexity threshold. This represents the maximum complexity threshold.

[0055] The optimized set of second feature points is obtained using the following formula:

[0056] In formula (11), This represents the optimized set of the second feature points. Indicates the total number of regions. Indicates the first The first region The coordinates of the feature points Indicates the first Frequency adjustment factor for each region Indicates the first The total number of candidate feature points in each region.

[0057] Adaptive parameter adjustments are used to dynamically increase the feature extraction frequency for high-complexity regions. For the eye region, due to its rich detail, the extraction frequency is increased from 10 times per second to 20 times per second, while the cheek region maintains the default frequency of 10 times per second. This adjustment generates a second set of feature points, increasing the number of feature points to approximately 6000, with even richer detail. Adaptive adjustment ensures more comprehensive feature capture in high-complexity regions while avoiding resource waste in low-complexity regions.

[0058] Step S240: Use a feature mapping algorithm to spatially integrate the key feature points in the second feature point set, map image information, depth information and thermal imaging details to a unified coordinate system, and generate an initial feature map containing feature distribution.

[0059] The initial feature map containing the feature distribution is obtained by the following formula:

[0060] In formula (12), Indicates the position in the initial feature map The characteristic distribution value at that location, This represents the total number of feature points in the second feature point set. Indicates the first The weights of each feature point Represents the Dirac function, and Indicates the first The coordinates of the feature points Indicates the first The intensity value of each feature point.

[0061] Feature mapping algorithms integrate the second set of feature points into a unified coordinate system. For example, image information is represented by two-dimensional pixel coordinates, depth information by three-dimensional point cloud coordinates, and thermal imaging is associated with temperature values ​​and two-dimensional coordinates. Through coordinate transformation algorithms, image information, depth information, and thermal imaging details are mapped to a unified three-dimensional coordinate system to generate an initial feature map. For example, eye feature points are represented in the coordinate system as (x, y, z, t), where x and y are planar positions, z is depth, and t is the temperature value. This mapping method facilitates subsequent analysis of the spatial relationships of features and improves data integration efficiency. In face recognition scenarios, the initial feature map can be used for subsequent identity verification or emotion analysis. The high-density feature points in the eye region ensure the capture of subtle expressions, depth information assists in stereo modeling, and thermal imaging data enhances the accuracy of liveness detection.

[0062] Furthermore, the AI ​​simulation method for facial cosmetic surgery proposed in this embodiment includes step S300: Step S310: Obtain segmentation information from the initial feature map, and use a region-growing-based segmentation algorithm to divide the initial feature map into regions to obtain the first region set.

[0063] The first region set is obtained using the following formula: (13) In formula (13), Indicates the first Each growth region Represents the pixels in the feature map. Represents the initial feature map. Represents pixels eigenvalues, Indicates the first The mean of each region, This represents the threshold for region growth. Represents pixels to seed point set distance, This indicates the maximum growth distance limit.

[0064] In face recognition scenarios, when obtaining segmentation information from the initial feature map, a region-growing-based segmentation algorithm is used for region division. The core idea of ​​this algorithm is to start from a seed point and gradually expand to neighboring pixels, merging them based on feature similarity such as color, texture, or depth values. Specifically, for a 1080p face image, the initial feature map contains pixel coordinates, depth values, and temperature values. High-density feature points in the eye region are selected as seed points, and a similarity threshold is set to a pixel grayscale difference of less than 10 or a depth value difference of less than 5 millimeters. The region-growing-based segmentation algorithm expands from the seed points, grouping neighboring pixels that meet the conditions into the same region, generating a first region set containing approximately 10 main regions, including the eyes, nose, and mouth. This method can effectively distinguish regions with significant feature differences, such as the texture differences between the eyes and cheeks.

[0065] Step S320: For the first set of regions, calculate the degree of boundary change of each region using a gradient analysis algorithm. If the degree of boundary change exceeds a preset threshold, it is identified as a high-change region, and the boundary change distribution is obtained.

[0066] The degree of boundary change for each region is calculated using the following formula:

[0067] In formula (14), Indicates the first The degree of boundary change in each region This represents the total number of pixels on the boundary of the region. Indicates the first on the boundary The intensity value of each pixel is used to measure boundary changes by calculating the average intensity difference between adjacent boundary pixels.

[0068] Set the criteria for determining high-variability regions: (15) In formula (15), Indicates the first The determination result of whether a region is a high-change region. Indicates the first The numerical values ​​of boundary change obtained from the calculation of each region This indicates a preset threshold value; the degree of change at the boundary is indicated by the numerical value. Exceeding the preset threshold The time is marked as 1, indicating a high-change region; otherwise, it is marked as 0.

[0069] For the first set of regions, a gradient analysis algorithm is used to calculate the degree of boundary change for each region. Gradient analysis focuses on abrupt changes in feature values; for example, the pixel grayscale gradient in the eye region might reach 50, while in the cheek region it might only be 20. A threshold of 30 is set; regions with a value higher than this, such as the eyes, are marked as high-change regions, while those with a value lower than this, such as the cheeks, are marked as low-change regions. The boundary change distribution is visualized as a gradient heatmap, intuitively showing the high-change characteristics of regions such as the eyes and nose, facilitating the focus on key regions during subsequent processing.

[0070] Step S330: Based on the boundary change distribution, use an adaptive mesh partitioning algorithm to increase the mesh density in high-change areas and maintain the default mesh density in low-change areas to obtain the second region set.

[0071] The second region set is obtained using the following formula: (16) In formula (16), Represents the second region set. and Indicates the first The coordinates of each grid node, Represents the set of natural numbers. Indicates the first Mesh density at each node This represents the grid density threshold.

[0072] Based on the boundary variation distribution, the adaptive mesh generation algorithm increases the mesh density in highly variable regions. In the eye region, due to the high gradient value, the mesh density is increased from 25 mesh points per square centimeter to 50, while the cheek region retains the default 25 mesh points. This adaptive partitioning generates a second set of regions containing more refined sub-regions, approximately 15 sub-regions. Refining the mesh ensures more accurate capture of feature points in highly variable regions, suitable for analyzing complex textures.

[0073] Step S340: Extract feature points from the clusters in the second region set, and use a feature clustering algorithm to perform regional clustering on the extracted feature points in the clusters to obtain the final feature subset set.

[0074] The final feature subset set is obtained using the following formula:

[0075] In formula (17), Represents the final subset of features. This represents the total number of clusters. Indicates the first The optimal representative feature point in each cluster Indicates the first A cluster, Representing feature points in a cluster, Represents other feature points within the same cluster. This represents the Euclidean distance between two feature points.

[0076] After extracting feature points from the second region set, a feature clustering algorithm is used for region-specific clustering. This algorithm is based on the spatial distance and attribute similarity between feature points. For example, using the K-means algorithm, feature points in the eye region are divided into three subsets based on texture and depth values, with each subset containing approximately 1000 feature points. The final feature subset set contains about 20 subsets, covering all key facial regions. This clustering method organizes feature points into structured subsets, facilitating subsequent analysis such as expression recognition or contour modeling.

[0077] Preferably, the AI ​​simulation method for facial cosmetic surgery proposed in this embodiment includes step S400 as follows: Step S410: Obtain the regional feature distribution from the feature subset, and use a stereomicroscope scanning algorithm to collect three-dimensional data of each region. If the tissue thickness of the region is lower than the preset threshold, increase the scanning frequency to obtain a high-resolution three-dimensional data point set.

[0078] Set the scanning frequency of the stereomicroscope using the following formula: (18) In formula (18), This indicates the scanning frequency of the stereomicroscope. Indicates the base scan frequency. Indicates the tissue thickness of the current region. This indicates the preset thickness threshold. This indicates the factor that increases the scanning frequency. Greater than 1.

[0079] The high-resolution 3D data point set is obtained through the following formula: (19) In formula (19), Represents a high-resolution set of three-dimensional data points. Indicates the first data points coordinate, Indicates the first data points coordinate, Indicates the first data points coordinate, Indicates the first The intensity value of each data point This represents the total number of 3D data points collected.

[0080] In 3D facial modeling, a stereomicroscope scanning algorithm is used to acquire 3D data when obtaining the regional feature distribution from a feature subset. Stereomicroscope scanning uses a high-precision optical system to capture the depth and surface information of the target area, generating 3D point cloud data. Specifically, for a feature subset of a 1080p facial image, including feature points in areas such as the eyes and nose, each feature point includes pixel coordinates and a depth value. During scanning, assuming the tissue thickness in the eye area is 2 mm, which is below a preset threshold of 5 mm, the scanning frequency is increased from 10 scans per second to 20 scans per second to capture finer depth variations. This high-frequency scanning generates a high-resolution 3D data point set containing approximately 5 million points, covering key facial areas and ensuring the accuracy of subsequent modeling.

[0081] Step S420: Based on the high-resolution three-dimensional data point set, an initial three-dimensional mesh is generated using a voxelization algorithm. If the boundary variation of the mesh cells in the initial three-dimensional mesh exceeds a preset threshold, the local mesh of the region is refined to obtain an optimized three-dimensional mesh.

[0082] The degree of boundary variation of a mesh cell is calculated using the following formula: (20) In formula (20), This indicates the magnitude of gradient change at the boundary of the grid cells. Functions representing the geometric or physical properties of mesh elements. Represents the third in three-dimensional space Each coordinate axis direction The attribute function is in the first position. The partial derivatives in each direction, when the gradient changes at the grid cell boundary. Local mesh refinement is triggered when the preset threshold is exceeded.

[0083] The recursive subdivision law of mesh size during local mesh refinement is described by the following formula: (twenty one) In formula (21), This indicates the new mesh size after local mesh refinement. Indicates the size of the original mesh cell. Indicates the number of refinement levels; by refining the original mesh according to 2... The grid is encrypted by dividing the area into equal parts using the power of the power.

[0084] Based on a high-resolution 3D data point set, a voxelization algorithm is used to generate an initial 3D mesh. The voxelization algorithm converts point cloud data into a regular voxel mesh, similar to dividing 3D space into small cubes. For example, for point cloud data of the nose region, setting the voxel size to 1 mm generates an initial mesh containing approximately 100,000 voxels. Gradient analysis is used to calculate the degree of boundary variation of the mesh cells. If the gradient value in the nose tip region reaches 60, exceeding a preset threshold of 40, then local mesh refinement is performed in that region, reducing the voxel size to 0.5 mm, forming an optimized 3D mesh containing approximately 150,000 voxels. This refinement ensures more accurate details in complex curved surface areas.

[0085] Step S430: Extract surface feature points from the optimized 3D mesh, and use the weighted least squares method to perform splicing weight allocation on the extracted surface feature points. If the tissue thickness of the region is lower than the preset threshold, the weight allocation ratio is increased to obtain a weighted feature point set.

[0086] Surface feature points with significant geometric characteristics are extracted by calculating the sum of squares of the principal curvatures at each point on the mesh surface.

[0087] In formula (22), Indicates the first A surface feature point, Represents the third in a 3D mesh A local area, and Representing points respectively The principal curvature value at that point.

[0088] The splicing weights are allocated using the following formula, which combines the inverse distance and the residual distribution:

[0089] In formula (23), Indicates the first Weighted least squares concatenation weights for each feature point This represents the total number of neighboring feature points. Representing feature points and neighboring points The distance between them Representing feature points The residual vector, This represents the mean of the residuals. This represents the Gaussian kernel parameters.

[0090] The following formula is used to implement an adaptive adjustment mechanism that increases the weight allocation ratio through an exponential function when the tissue thickness is below a preset threshold:

[0091] In formula (24), This represents the adjusted weight value. This represents the original weight values. Indicates the weight enhancement coefficient. Indicates a thickness-sensitive parameter. Indicates the region The tissue thickness value, e Represents the natural constant.

[0092] After extracting surface feature points from the optimized 3D mesh, weighted least squares (WLS) is used to assign stitching weights. WLS optimizes stitching accuracy by analyzing the spatial relationships between feature points. Specifically, for feature points in the eye region, if the tissue thickness is less than 5 mm, the weight allocation ratio is increased from the default 0.3 to 0.5, generating approximately 20,000 weighted feature points. This weighting method highlights the feature contributions of thin tissue regions, helping to improve the model's detail representation.

[0093] Step S440: Based on the weighted feature point set, a preliminary three-dimensional model is generated using a volume rendering algorithm. If the surface curvature of the preliminary three-dimensional model exceeds a preset threshold, the surface curvature exceeding the preset threshold is locally smoothed to obtain a high-precision three-dimensional model.

[0094] The curvature of each point on the surface of the 3D model is calculated using the following formula to determine whether it exceeds a preset threshold: (25) In formula (25), Indicates the position of the surface of the 3D model The curvature value at that point, Represents the gradient operator, Indicates surface position The unit normal vector at that location, Represents the position vector on the surface; The following formula is used to locally smooth surface regions with curvature exceeding a threshold: (26) In formula (26), Represents the vertices after smoothing. new location, Represents vertices The original location, This indicates the smoothing intensity control parameter. Represents vertices The set of neighboring vertices, Representing the neighboring vertices The weighting coefficients, Representing the neighboring vertices The location.

[0095] Based on a weighted set of feature points, a volume rendering algorithm is used to generate a preliminary 3D model. The volume rendering algorithm forms a continuous surface model by rendering voxel data. For example, for the mouth region, the calculated surface curvature of the preliminary model is 0.8, which is higher than the preset threshold of 0.5. Therefore, local smoothing is performed on this region, using mean filtering to reduce the curvature to 0.4, generating a high-precision 3D model. This smoothing process ensures natural transitions on the model surface, making it suitable for subsequent applications such as animation modeling or medical analysis.

[0096] Furthermore, the AI ​​simulation method for facial cosmetic surgery proposed in this embodiment includes step S500: Step S510: Obtain regional geometric data from the high-precision 3D model, and use a stereomicroscope scanning algorithm to extract local feature distribution. If the curvature of the local features exceeds a preset threshold, increase the sampling frequency to obtain a high-resolution geometric dataset.

[0097] The curvature value of the local geometric feature is calculated using the following formula to determine whether it exceeds a preset threshold: (27) In formula (27), This represents the Gaussian curvature of the surface of the 3D model at the parametric coordinates (u, v). The height function represents the height of a high-precision 3D model. and Representing functions respectively For parameters and The first-order partial derivative, , , These represent the corresponding second-order partial derivatives.

[0098] When the local curvature exceeds a preset threshold, the sampling frequency is automatically increased using the following formula: (28) In formula (28), Indicates the location Adaptive sampling frequency at the location, Indicates the basic sampling frequency. Indicates the sampling enhancement coefficient. This represents the local curvature value at the current location. This represents the preset curvature threshold. This indicates the sampling adjustment parameters.

[0099] The local geometric feature distribution can be extracted from microscopic scanning data using the following formula: (29) In formula (29), This represents the local feature response extracted by the stereomicroscope scanning algorithm. This represents the input microscope image data. Represents the Laplace operator. The standard deviation is expressed as Gaussian kernel function, This represents the convolution operation. Indicates direction as Directional filter.

[0100] In 3D facial modeling, when acquiring regional geometric data from a high-precision 3D model, a stereomicroscopy scanning algorithm is used to extract local feature distributions. Stereomicroscopy captures geometric information of facial regions, such as the curvature of the corners of the eyes or the nostrils, using a high-resolution optical system. Specifically, for a geometric dataset of a 1080p facial image, assuming the curvature of the nose tip region is 0.7, exceeding a preset threshold of 0.5, the sampling frequency is increased from 15 times per second to 25 times per second, generating a high-resolution dataset containing approximately 6 million geometric data points. This high-frequency sampling captures finer surface variations, ensuring the accuracy of subsequent rendering.

[0101] Step S520: Based on the high-resolution geometric dataset, use the stereo rendering algorithm to generate initial lighting and shadow data, and adjust the rendering brightness in combination with preset effect parameters to obtain a preliminary lighting and shadow rendering image.

[0102] The adjusted rendering brightness is obtained using the following formula:

[0103] In formula (30), This indicates the adjusted brightness value. Indicates the original brightness value. This represents the base brightness adjustment factor. This represents the weighting coefficient of the effect parameter. This indicates the total number of preset effect parameters. Indicates the first The weight of each effect parameter, Indicates the first The contribution of each preset effect to a pixel position.

[0104] Based on a high-resolution geometric dataset, a stereoscopic rendering algorithm is used to generate initial lighting and shadow data. This algorithm simulates the interaction between light and geometric surfaces to create preliminary lighting and shadow effects. For example, for the mouth area, setting the rendering brightness parameter to 80% generates initial lighting and shadow data that highlights the lip area. If the user requests enhanced highlight effects, the brightness is adjusted to 90% to create a more three-dimensional lighting and shadow rendering. This adjustment improves visual realism and is suitable for virtual reality applications.

[0105] Step S530: Extract surface texture points from the preliminary lighting and shadow rendering map. If the user's requirements include height modification, the weighted least squares method is used to assign priority weights to the texture points to obtain a weighted texture point set.

[0106] Extract surface texture points from the initial lighting and shadow rendering using the following formula:

[0107] In formula (31), Indicates the first Extraction values ​​of surface texture points Represents pixels The neighborhood set, Represents pixels Light intensity value, Represents pixels The illumination gradient vector at that location.

[0108] Assign priority weights to texture points based on height modification requirements using the following formula:

[0109] In formula (32), Indicates the first The priority weight value of each texture point Indicates the first The required intensity of height modification for each texture point Indicates the total number of texture points. Indicates a high sensitivity parameter. Represents the regularization parameter. Indicates the first The required intensity of height modification for each texture point.

[0110] The mathematical representation of the final generated weighted texture point set is defined by the following formula: (33) In formula (33), Represents a weighted set of texture points. , , They represent the first The three-dimensional spatial coordinates of a texture point Indicates the first The weight value corresponding to each texture point. This represents the total number of weighted texture points.

[0111] When extracting surface texture points from the initial lighting and shadow rendering, if the user's requirements include height modification, such as highlighting eye texture, a weighted least squares method is used to assign priority weights. Specifically, for texture points in the eye region, assuming their height value is less than 2 millimeters, the weight is increased from the default 0.2 to 0.4, generating approximately 30,000 weighted texture points. This weighting method can highlight subtle texture features, improve the model's detail representation, and is suitable for refined modeling needs.

[0112] Step S540: Based on the weighted texture point set, a volume rendering algorithm is used to generate an effect simulation image. If the smoothness of the effect simulation image is lower than a preset threshold, local smoothing optimization is performed on the relevant areas to obtain the final rendering effect image.

[0113] The final rendered image is obtained using the following formula:

[0114] In formula (34), Indicates the pixel coordinates of the rendered image The final color value at that location, This represents the total number of weighted texture points. Indicates the first The weight value of each texture point Indicates the first The texture contribution of each texture point at a pixel location. Indicates the first The transparency of each texture point Represents the volume density function. Indicates from the viewpoint to the first The distance between texture points e Represents the natural constant.

[0115] Based on a weighted set of texture points, a volume rendering algorithm is used to generate a simulated effect image. The volume rendering algorithm creates a continuous surface effect by rendering texture points. For example, for the chin area, if the smoothness value of the simulated image is 0.3, which is lower than the preset threshold of 0.6, local smoothing optimization is performed on this area. Mean filtering is used to increase the smoothness to 0.5, generating the final rendered image. This smoothing process reduces surface abruptness and is suitable for medical visualization or animation modeling.

[0116] Please see Figure 2 This embodiment provides an AI simulation system for facial cosmetic surgery, used to execute the aforementioned AI simulation method for facial cosmetic surgery. It includes a multidimensional dataset acquisition module 10, an initial feature map acquisition module 20, a feature subset acquisition module 30, a high-precision 3D model acquisition module 40, and an effect simulation image acquisition module 50. The multidimensional dataset acquisition module 10 acquires multi-source biometric data from a data acquisition device, classifies the feature complexity of the multi-source biometric data, and assigns high-performance nodes to process fine details of specific regions within the multi-source biometric data to obtain a classified multidimensional dataset. The multi-source biometric data includes image information, depth information, and thermal imaging details. The initial feature map acquisition module 20 extracts key feature points from the classified multidimensional dataset using a feature extraction algorithm, adaptively adjusting parameters based on the complexity of different regions. If the feature distribution in a specific region is dense, the extraction frequency is increased to obtain an initial feature map containing the feature distribution. The feature subset acquisition module 30 is used to obtain segmentation information from the initial feature map, divide different regions using a segmentation algorithm, adjust the segmentation granularity for structural differences, and refine the mesh if the boundary of the segmented region changes significantly to obtain a feature subset of the region. The high-precision 3D model acquisition module 40 is used to generate a model using a 3D reconstruction algorithm based on the feature subset of the region, dynamically adjust the stitching weights for the scanning parameters of each region, and increase the stitching accuracy if the tissue thickness of a specific region is less than a threshold to obtain a high-precision 3D model. The effect simulation image acquisition module 50 is used to obtain rendering data from the high-precision 3D model, perform simulation rendering in combination with preset effect parameters, adjust the lighting and texture according to user needs, and prioritize rendering the relevant regions if the user needs include height modification to obtain the final effect simulation image.

[0117] This embodiment provides an AI simulation method and system for facial cosmetic surgery. Compared with existing technologies, it proposes a solution for dynamically adjusting processing strategies to address the complexity differences in image, depth information, and thermal imaging details within multi-source biometric data. For specific regions with high feature complexity, high-performance nodes are allocated for processing, and an adaptive feature extraction algorithm is used to dynamically adjust the extraction frequency and segmentation granularity, generating a feature subset containing fine-grained feature distributions. Based on this, a 3D reconstruction algorithm is used to dynamically adjust the stitching weights, particularly improving stitching accuracy in thinner tissue regions, generating a high-precision 3D model. Finally, user-driven rendering parameter adjustments prioritize rendering regions with height modifications, resulting in a realistic simulation image. This embodiment significantly improves the modeling accuracy and rendering efficiency of complex biometric data through adaptive parameter adjustment and refined processing, making it suitable for scenarios such as medical imaging and biometric recognition.

[0118] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. An AI simulation method for facial cosmetic surgery, characterized in that, Includes the following steps: S100. Acquire multi-source biometric data from the data acquisition device, classify the feature complexity of the multi-source biometric data, and if the multi-source biometric data contains fine details of a specific region, allocate high-performance nodes for processing to obtain a classified multidimensional dataset, wherein the multi-source biometric data contains image information, depth information and thermal imaging details. S200. Based on the classified multidimensional dataset, a feature extraction algorithm is used to extract key feature points. The parameters are adaptively adjusted according to the complexity of different regions. If the feature distribution in a specific region is dense, the extraction frequency is increased to obtain an initial feature map containing the feature distribution. S300. Obtain segmentation information from the initial feature map, use a segmentation algorithm to divide different regions, adjust the segmentation granularity according to structural differences, and if the boundary of the segmented region changes significantly, refine the grid to obtain feature subsets of the sub-regions. S400. Based on the feature subsets of the sub-regions, a three-dimensional reconstruction algorithm is used to generate a model. The stitching weights are dynamically adjusted for the scanning parameters of each region. If the tissue thickness of a specific region is less than the threshold, the stitching accuracy is improved to obtain a high-precision three-dimensional model. S500: Obtain rendering data from the high-precision 3D model, perform simulation rendering in combination with preset effect parameters, adjust lighting and texture according to user requirements, and if the user requirements include height modification, prioritize rendering the relevant areas to obtain the final effect simulation image.

2. The AI ​​simulation method for facial cosmetic surgery as described in claim 1, characterized in that, Step S100 includes: S110. Receive multi-source biometric data containing image information, depth information, and thermal imaging details from the data acquisition device, and perform standardized processing on the multi-source biometric data through a preset sensor interface to obtain a multi-source data set in a unified format. A unified format for a multi-source dataset is derived using the following formula: in, This represents a multi-source data set in a unified format. Indexes representing three types of biometric data: image information, depth information, and thermal imaging details. Indicates the first Weighting coefficients for different data types Indicates the first Standardized processing functions for sensor interfaces corresponding to various data types. Indicates the first A variety of original biometric data sources, among which When the value is 1, it corresponds to the image information. When the value is 2, it corresponds to depth information. A value of 3 corresponds to thermal imaging details; S120. For the multi-source data set, a feature extraction algorithm is used to quantitatively analyze the feature complexity of the image information, depth information and thermal imaging details. If the feature complexity exceeds a preset threshold, it is determined to be a high-complexity feature, and the feature complexity level after classification is obtained. The feature complexity of the image information is derived using the following formula: in, The quantization value representing the feature complexity of image information. and These represent the height and width of the image in pixels, respectively. Indicates position Pixel value at that location, and These respectively indicate that the position is in direction and The gradient value in the direction is used to quantify the complexity of image features by calculating the average gradient magnitude across the entire image. The feature complexity of the depth information is derived by the following formula: in, The quantization value representing the feature complexity of depth information. This represents the total number of regions in the depth image. Indicates the first The variance of depth values ​​within each region block Indicates the first The mean depth value within each region block This represents a small constant to prevent division by zero, and the complexity of the depth feature is evaluated by calculating the coefficient of variation of the depth change of each region block; Set the criteria for determining the feature complexity level after classification: in, This indicates the level of feature complexity after classification. This represents the numerical value of feature complexity calculated using the feature extraction algorithm. This represents a preset complexity threshold. When the feature complexity exceeds the threshold, it is judged as a high-complexity feature; otherwise, it is a low-complexity feature, thus realizing the classification of complexity levels of features from multiple sources. S130. Based on the feature complexity level, if it contains fine details of a specific region, the high-complexity feature is distributed to high-performance nodes for processing using a load balancing algorithm to obtain an optimized multidimensional feature dataset. The quality assessment value of the optimized multidimensional feature dataset is obtained using the following formula: in, This represents the quality assessment value of the optimized multidimensional feature dataset. This represents the total number of features after processing. Indicates the first The importance factors of each feature Indicates the first Information entropy of each feature Indicates the first The processing quality coefficient of each feature; S140. The image information, depth information and thermal imaging details in the multidimensional feature dataset are integrated using a data fusion algorithm to generate a multidimensional dataset; Multidimensional data sets are derived using the following formula: in, This represents the integrated cube. A numerical vector representing image information. A numerical vector representing depth information. Numerical vectors representing details in thermal imaging. The standard deviation representing image information The standard deviation representing depth information The standard deviation of thermal imaging information.

3. The AI ​​simulation method for facial cosmetic surgery as described in claim 1, characterized in that, Step S200 includes: S210. Obtain multidimensional data containing image information, depth information and thermal imaging details from the classified multidimensional dataset, process the multidimensional data through a preset feature extraction algorithm, extract key feature points, and obtain a first feature point set. The first set of feature points is obtained by the following formula: in, Indicates the first The fused multidimensional feature values, Indicates position Image information intensity at that location Indicates position Depth information value at that location, Indicates position Thermal imaging detail data at the location, , , These represent the fusion weight coefficients for the three data types: image information, depth information, and thermal imaging details, respectively. S220. For the first set of feature points, a quantitative evaluation method is used to analyze the feature point density of different regions. If the feature point density of a region exceeds a preset threshold, the region is determined to be a high-complexity region, and the region complexity distribution is obtained. The density of feature points in a region is obtained using the following formula: in, Indicates the first Feature point density of each region Indicates the first The number of feature points in a region Indicates the first The size of each region; Define the conditions used to determine the complexity category of a region: in, Indicates the first Complexity identifier for each region This represents the preset density threshold, when the first... Feature point density of each region Exceeding the preset density threshold The time marker is designated as a high-complexity region; S230. Based on the distribution of regional complexity, the feature extraction frequency of high-complexity regions is dynamically increased through an adaptive parameter adjustment algorithm, while the default frequency is maintained for low-complexity regions, to obtain an optimized set of second feature points. The feature extraction frequency for high-complexity regions is dynamically increased using the following formula: in, Indicates the first Feature extraction frequency of each region This indicates the default base frequency. Indicates the frequency adjustment factor. Indicates the exponential growth parameter. Indicates the first Complexity assessment value for each region This represents the minimum complexity threshold. This represents the maximum complexity threshold; The optimized set of second feature points is obtained using the following formula: in, This represents the optimized set of the second feature points. Indicates the total number of regions. Indicates the first The first region The coordinates of the feature points Indicates the first Frequency adjustment factor for each region Indicates the first The total number of candidate feature points in each region; S240. Use a feature mapping algorithm to spatially integrate the key feature points in the second feature point set, map image information, depth information and thermal imaging details to a unified coordinate system, and generate an initial feature map containing feature distribution. The initial feature map containing the feature distribution is obtained by the following formula: in, Indicates the position in the initial feature map The characteristic distribution value at that location, This represents the total number of feature points in the second feature point set. Indicates the first The weights of each feature point Represents the Dirac function, and Indicates the first The coordinates of the feature points Indicates the first The intensity value of each feature point.

4. The AI ​​simulation method for facial cosmetic surgery as described in claim 1, characterized in that, Step S300 includes: S310. Obtain segmentation information from the initial feature map, and use a region-growing-based segmentation algorithm to divide the initial feature map into regions to obtain a first region set; The first region set is obtained using the following formula: in, Indicates the first Each growth region Represents the pixels in the feature map. Represents the initial feature map. Represents pixels eigenvalues, Indicates the first The mean of each region, This represents the threshold for region growth. Represents pixels to seed point set distance, Indicates the maximum growth distance limit; S320. For the first set of regions, the degree of boundary change of each region is calculated by gradient analysis algorithm. If the degree of boundary change exceeds a preset threshold, it is determined to be a high-change region, and the boundary change distribution is obtained. The degree of boundary change for each region is calculated using the following formula: in, Indicates the first The degree of boundary change in each region This represents the total number of pixels on the boundary of the region. Indicates the first on the boundary The intensity value of each pixel is used to measure boundary changes by calculating the average intensity difference between adjacent boundary pixels. Set the criteria for determining high-variability regions: in, Indicates the first The determination result of whether a region is a high-change region. Indicates the first The numerical values ​​of boundary change obtained from the calculation of each region This indicates a preset threshold value; the degree of change at the boundary is indicated by the numerical value. Exceeding the preset threshold A time interval is marked as 1, indicating a high-change region; otherwise, it is marked as 0. S330. Based on the boundary change distribution, an adaptive grid partitioning algorithm is used to increase the grid density in high-change areas and maintain the default grid density in low-change areas to obtain a second set of regions. The second region set is obtained using the following formula: in, Represents the second region set. and Indicates the first The coordinates of each grid node, Represents the set of natural numbers. Indicates the first Mesh density at each node Indicates the grid density threshold; S340. Extract feature points from the clusters in the second region set, and use a feature clustering algorithm to perform regional clustering on the extracted feature points in the clusters to obtain the final feature subset set. The final feature subset set is obtained using the following formula: in, Represents the final subset of features. This represents the total number of clusters. Indicates the first The optimal representative feature point in each cluster Indicates the first A cluster, Representing feature points in a cluster, Represents other feature points within the same cluster. This represents the Euclidean distance between two feature points.

5. The AI ​​simulation method for facial cosmetic surgery as described in claim 1, characterized in that, Step S400 includes: S410. Obtain the regional feature distribution from the feature subset, and use a stereomicroscope scanning algorithm to collect three-dimensional data of each region. If the tissue thickness of the region is lower than a preset threshold, increase the scanning frequency to obtain a high-resolution three-dimensional data point set. Set the scanning frequency of the stereomicroscope: in, This indicates the scanning frequency of the stereomicroscope. Indicates the base scan frequency. Indicates the tissue thickness of the current region. This indicates the preset thickness threshold. This indicates the scanning frequency is increased by a factor and Greater than 1; The high-resolution 3D data point set is obtained through the following formula: in, Represents a high-resolution set of three-dimensional data points. Indicates the first data points coordinate, Indicates the first Data points coordinate, Indicates the first data points coordinate, Indicates the first The intensity value of each data point This represents the total number of 3D data points collected. S420. Based on the set of high-resolution three-dimensional data points, an initial three-dimensional mesh is generated using a voxelization algorithm. If the boundary variation of the mesh cells of the initial three-dimensional mesh exceeds a preset threshold, the local mesh of the region is refined to obtain an optimized three-dimensional mesh. S430. Extract surface feature points from the optimized three-dimensional mesh, and use the weighted least squares method to perform splicing weight allocation on the extracted surface feature points. If the tissue thickness of the region is lower than a preset threshold, increase the weight allocation ratio to obtain a weighted feature point set. S440. Based on the weighted feature point set, a preliminary three-dimensional model is generated using a volume rendering algorithm. If the surface curvature of the preliminary three-dimensional model exceeds a preset threshold, the surface curvature exceeding the preset threshold is locally smoothed to obtain a high-precision three-dimensional model.

6. The AI ​​simulation method for facial cosmetic surgery as described in claim 5, characterized in that, In step S420, the degree of boundary variation of the mesh element is calculated using the following formula: in, This indicates the magnitude of gradient change at the boundary of the grid cells. Functions representing the geometric or physical properties of mesh elements. Represents the third in three-dimensional space Each coordinate axis direction The attribute function is in the first position. The partial derivatives in each direction, when the gradient changes at the grid cell boundary. Local mesh refinement is triggered when the preset threshold is exceeded; The recursive subdivision law of mesh size during local mesh refinement is described by the following formula: in, This indicates the new mesh size after local mesh refinement. Indicates the size of the original mesh cell. Indicates the number of refinement levels; by refining the original mesh according to 2... The grid is encrypted by dividing the area into equal parts using the power of the power.

7. The AI ​​simulation method for facial cosmetic surgery as described in claim 6, characterized in that, In step S430, surface feature points with significant geometric characteristics are extracted by calculating the sum of squares of the principal curvatures of each point on the mesh surface: in, Indicates the first A surface feature point, Represents the third in a 3D mesh A local area, and Representing points respectively The principal curvature value at that point; The splicing weights are allocated using the following formula, which combines the inverse distance and the residual distribution: in, Indicates the first Weighted least squares concatenation weights for each feature point This represents the total number of neighboring feature points. Representing feature points and neighboring points The distance between them Representing feature points The residual vector, This represents the mean of the residuals. Indicates the Gaussian kernel parameters; The following formula is used to implement an adaptive adjustment mechanism that increases the weight allocation ratio through an exponential function when the tissue thickness is below a preset threshold: in, This represents the adjusted weight value. This represents the original weight values. Indicates the weight enhancement coefficient. Indicates a thickness-sensitive parameter. Indicates the region The tissue thickness value, where e represents the natural constant.

8. The AI ​​simulation method for facial cosmetic surgery as described in claim 7, characterized in that, In step S440, the curvature of each point on the surface of the 3D model is calculated using the following formula to determine whether it exceeds a preset threshold: in, Indicates the position of the surface of the 3D model The curvature value at that point, Represents the gradient operator, Indicates surface position The unit normal vector at that location, Represents the position vector on the surface; The following formula is used to locally smooth surface regions with curvature exceeding a threshold: in, Represents the vertices after smoothing. new location, Represents vertices The original location, This indicates the smoothing intensity control parameter. Represents vertices The set of neighboring vertices, Representing the neighboring vertices The weighting coefficients, Representing the neighboring vertices The location.

9. The AI ​​simulation method for facial cosmetic surgery as described in claim 1, characterized in that, Step S500 includes: S510. Obtain regional geometric data from the high-precision three-dimensional model, and extract local feature distribution using a stereomicroscope scanning algorithm. If the curvature of the local features exceeds a preset threshold, increase the sampling frequency to obtain a high-resolution geometric dataset. S520. Based on the high-resolution geometric dataset, the initial lighting and shadow data is generated using a stereo rendering algorithm. The rendering brightness is adjusted in combination with preset effect parameters to obtain a preliminary lighting and shadow rendering image. S530. Extract surface texture points from the preliminary lighting and shadow rendering map. If the user requirement includes height modification, use the weighted least squares method to assign priority weights to the texture points to obtain a weighted texture point set. S540. Based on the weighted texture point set, a volume rendering algorithm is used to generate an effect simulation image. If the smoothness of the effect simulation image is lower than a preset threshold, local smoothing optimization is performed on the relevant areas to obtain the final rendering effect image.

10. An AI simulation system for facial cosmetic surgery, used to execute the AI ​​simulation method for facial cosmetic surgery as described in any one of claims 1 to 9, characterized in that, include: The multidimensional dataset acquisition module (10) is used to acquire multi-source biometric data from the data acquisition device, classify the feature complexity of the multi-source biometric data, and if the multi-source biometric data contains fine details of a specific region, then allocate high-performance nodes for processing to obtain the classified multidimensional dataset, wherein the multi-source biometric data contains image information, depth information and thermal imaging details. The initial feature map acquisition module (20) is used to extract key feature points based on the classified multidimensional dataset using a feature extraction algorithm. The parameters are adaptively adjusted according to the complexity of different regions. If the feature distribution in a specific region is dense, the extraction frequency is increased to obtain an initial feature map containing the feature distribution. The feature subset acquisition module (30) is used to acquire segmentation information from the initial feature map, divide different regions using a segmentation algorithm, adjust the segmentation granularity according to structural differences, and refine the grid if the boundary of the segmented region changes significantly to obtain the feature subset of the region. The high-precision three-dimensional model acquisition module (40) is used to generate a model based on the feature subsets of the sub-regions using a three-dimensional reconstruction algorithm, dynamically adjust the stitching weights of the scanning parameters of each region, and improve the stitching accuracy if the tissue thickness of a specific region is less than the threshold, thereby obtaining a high-precision three-dimensional model. The effect simulation image acquisition module (50) is used to acquire rendering data from the high-precision 3D model, perform simulation rendering in combination with preset effect parameters, adjust the lighting and texture according to user requirements, and if the user requirements include height modification, the relevant area is rendered first to obtain the final effect simulation image.

Images