Oral porcelain restoration color prediction system, construction method and prediction method

Abstract

The present application relates to the technical field of oral prosthetics, in particular to a porcelain dental prosthesis color prediction system, a construction method and a prediction method, comprising: a multi-modal image acquisition module for acquiring high-definition color images of natural teeth in the oral cavity, three-dimensional scanning images of tooth preparation surfaces and microstructure images of porcelain prosthesis substrates, and synchronously acquiring environmental light parameters; an image recognition processing module for processing original data, outputting natural tooth color space feature vectors, tooth preparation surface morphology parameters and porcelain layer substrate micro-pore distribution parameters; an intelligent color prediction module for constructing a deep learning model, combining an environmental light correction algorithm to predict the final presented color data of the prosthesis; a parameter output and optimization module for generating a color deviation report, outputting porcelain layer thickness adjustment suggestions and firing process parameter optimization schemes, and supporting visual display of the prediction effect. Thus, the problems of one-sided prediction dimension and subjective optimization in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of oral restoration technology, specifically to a color prediction system for dental porcelain restorations, its construction method, and prediction method. Background Technology

[0002] Due to their excellent biocompatibility and aesthetics, porcelain restorations have become a core restorative solution for dental diseases such as tooth defects and discoloration. The color matching between the restoration and the natural tooth directly determines the aesthetics of the restoration and patient satisfaction. In clinical practice, color matching of restorations must simultaneously consider the color characteristics of the natural tooth, the morphological differences after tooth preparation, the optical properties of the porcelain substrate, and the influence of ambient light during treatment, which is a technical challenge in the field of dental restorations. Traditional color matching methods mostly rely on visual comparison with shade guides or simple digital acquisition, which can only obtain color information from a limited area of ​​the natural tooth surface. They cannot capture the color gradation patterns of the labial, lingual, and proximal surfaces of the tooth, and do not consider the influence of the microscopic porosity distribution and particle aggregation of porcelain powder on light reflection and transmission, resulting in insufficient comprehensiveness and accuracy of color acquisition. At the same time, fluctuations in color temperature and illuminance of the treatment environment can directly interfere with color judgment, and traditional methods lack effective ambient light correction mechanisms, further exacerbating the color deviation between the restoration and the natural tooth.

[0003] Current color prediction technologies have significant limitations: Firstly, most prediction models are based on single color features and do not integrate the correlation between morphological parameters such as the area and curvature of the prepared tooth surface and the thickness of the porcelain layer and the firing process, making it difficult to accurately simulate the color presentation of the porcelain layer under different tooth morphologies. Secondly, optimizing process parameters relies on the doctor's subjective experience, requiring repeated trial firings to adjust the porcelain layer thickness and firing temperature curves, which not only prolongs the restoration period but also increases material waste and medical costs. These shortcomings lead to a high rate of color mismatch in restorations in clinical practice, requiring some patients to undergo secondary restorations, affecting the treatment experience and reducing treatment efficiency. This problem is particularly prominent in primary healthcare institutions due to a lack of professional color matching personnel and precise tools. There is an urgent need for an integrated system that can integrate multi-dimensional data, intelligently predict colors, and optimize process parameters to improve the color matching accuracy and clinical treatment efficiency of dental porcelain restorations. Summary of the Invention

[0004] This application provides a color prediction system, construction method, and prediction method for dental porcelain restorations to address the problems of limited prediction dimensions and subjective optimization in existing technologies.

[0005] The first aspect of this application provides a color prediction system for dental porcelain restorations, comprising: a multimodal image acquisition module, an image recognition and processing module, an intelligent color prediction module, and a parameter output and optimization module. The multimodal image acquisition module acquires high-resolution color images of natural teeth, three-dimensional scan images of the prepared tooth surface, and microstructural images of the porcelain restoration substrate, while simultaneously recording ambient light parameters. The image recognition and processing module processes the raw data and outputs a natural tooth color space feature vector, morphological parameters of the prepared tooth surface, and micropore distribution parameters of the porcelain substrate. The intelligent color prediction module constructs a deep learning model that integrates image features and the optical properties of the porcelain layer. It inputs the color feature vector, morphological parameters, micropore distribution parameters, and a preset porcelain layer thickness parameter, and combines this with an ambient light correction algorithm to predict the final color data of the dental porcelain restoration. The parameter output and optimization module compares the predicted color data with the color features of natural teeth, generates a color deviation report, outputs suggestions for porcelain layer thickness adjustment and optimization schemes for firing process parameters, and supports visual display of the prediction results.

[0006] Preferably, the multimodal image acquisition module includes an intraoral high-definition imaging unit, a three-dimensional structure scanning unit, a substrate microscopic imaging unit, and an ambient light monitoring unit. The intraoral high-definition imaging unit uses a macro camera with a color calibration card to acquire high-definition RGBB images of the labial, lingual, and proximal surfaces of natural teeth. The three-dimensional structure scanning unit acquires three-dimensional point cloud data of the prepared tooth surface through structured light scanning. The substrate microscopic imaging unit uses microscopic imaging technology to acquire microscopic structural images of porcelain powder before sintering and identify particle size and porosity distribution. The ambient light monitoring unit is used to acquire color temperature and illuminance data in the oral cavity and treatment environment in real time.

[0007] Preferably, the image recognition processing module includes an image processing unit, a color feature extraction unit, and a structural parameter recognition unit. The image processing unit is used to eliminate image interference caused by saliva reflection in the oral cavity and the shooting angle through image denoising, white balance correction, and shadow removal algorithms. The color feature extraction unit is used to segment the natural tooth region from the background using a CNN image recognition algorithm, extract the CIELab value of each pixel, and generate a natural tooth color distribution heatmap and an average color feature vector. The structural parameter recognition unit is used to extract the area, curvature, and edge morphology parameters of the tooth preparation surface from the three-dimensional scan image using image segmentation and edge detection algorithms, and to identify the porosity and particle aggregation state parameters of the porcelain substrate from the microstructure image.

[0008] Preferably, the intelligent color prediction module includes a feature fusion unit, a deep learning prediction model unit, and an ambient light correction unit. The feature fusion unit nonlinearly fuses the color feature vectors and structural parameters obtained from image recognition with the porcelain layer thickness and refractive index optical parameters to construct a multidimensional input feature matrix. The deep learning prediction model unit, based on a ResNet and Transformer hybrid architecture, is trained using clinically collected natural teeth and restoration color matching samples, and outputs the CIELab prediction value and color uniformity assessment of the restoration under target ambient light. The ambient light correction unit corrects the predicted color data based on ambient light monitoring data using a color temperature compensation algorithm, eliminating color deviations under different treatment environments.

[0009] Preferably, the parameter output and optimization module includes a color comparison unit, a process parameter generation unit, and a visualization unit. The color comparison unit is used to calculate the color difference between the predicted color and the natural tooth color. The process parameter generation unit is used to output thickness adjustment parameters for the bottom, middle, and top layers of the porcelain layer, as well as an optimization scheme for the firing temperature curve, based on the color deviation data. The visualization unit is used to overlay the predicted restoration color onto the patient's oral cavity scan image using AR, allowing doctors and patients to intuitively view the restoration effect.

[0010] The second aspect of this application provides a method for constructing a color prediction system for dental ceramic restorations, comprising: configuring multimodal image acquisition hardware including an intraoral high-definition imaging unit, a three-dimensional structural scanning unit, a substrate microscopic imaging unit, and an ambient light monitoring unit; building an image recognition processing flow integrating denoising correction, CNN feature extraction, and edge detection algorithms; constructing an intelligent color prediction model based on a hybrid ResNet and Transformer architecture and incorporating an ambient light correction algorithm; designing a parameter output and optimization device including color comparison, layer thickness and firing process parameter generation, and AR visualization functions; training the model with clinical samples and verifying the coherence of data flow and prediction accuracy of each device to complete the system construction.

[0011] The third aspect of this application provides a method for predicting the color of dental porcelain restorations, comprising: acquiring multimodal images of natural teeth in the oral cavity, microscopic images of the porcelain restoration substrate, and ambient light parameters; extracting natural tooth color feature vectors, tooth preparation surface morphology parameters, and porcelain layer microstructure parameters based on the multimodal images of natural teeth in the oral cavity, the microscopic images of the porcelain restoration substrate, and the ambient light parameters using an image recognition algorithm; inputting the natural tooth color feature vectors, tooth preparation surface morphology parameters, and porcelain layer microstructure parameters into a pre-trained intelligent color prediction model, and combining them with an ambient light correction algorithm to obtain predicted color data for the restoration; The predicted color data is compared with the natural tooth color to generate a color deviation report and optimized parameters for porcelain layer thickness and firing process, which are then displayed using AR.

[0012] A fourth aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the dental porcelain restoration color prediction method as described in the above embodiments.

[0013] A fifth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the dental porcelain restoration color prediction method as described in the above embodiments.

[0014] A sixth aspect of this application provides a computer program product, including a computer program or instructions, for implementing the dental porcelain restoration color prediction method as described in the above embodiments.

[0015] Therefore, this application has the following beneficial effects: This application embodiment simultaneously acquires high-definition color images of natural teeth, 3D scan images of the prepared tooth surface, and microstructure images of the porcelain substrate through a multimodal image acquisition module, while simultaneously recording ambient light parameters. This overcomes the limitations of traditional single-acquisition methods, enabling comprehensive data capture from the multi-faceted color gradations of natural teeth and tooth morphological features to the distribution of micropores in the porcelain layer. Combined with the image recognition processing module's noise reduction, CNN color feature extraction, and edge detection algorithms, saliva reflection and angular interference are eliminated, generating natural tooth color space feature vectors and tooth structure parameters, thus improving the comprehensiveness and accuracy of data acquisition. Furthermore, an intelligent color prediction module constructs a hybrid ResNet and Transformer architecture model, integrating color features... By incorporating morphological parameters, porcelain layer optical properties, and ambient light correction algorithms, this approach avoids the limitations of single-feature modeling. It accurately simulates the light reflection and transmission effects of the porcelain layer under different tooth morphologies, reducing the color difference between the predicted color of the restoration and the natural tooth, thus improving color matching naturalness. Through parameter output and optimization modules, it automatically compares the predicted color with the natural tooth color, generating a quantitative deviation report and optimization schemes for layered porcelain layer thickness and firing temperature curves. This replaces traditional experience-based subjective adjustments, reducing porcelain powder loss and prolonged restoration cycles caused by repeated firing trials. Furthermore, AR visualization overlays the predicted effect onto the oral cavity scene, allowing doctors and patients to intuitively confirm the restoration effect, improving clinical diagnostic efficiency and ensuring a high degree of color matching between the restoration and the natural tooth. Therefore, it solves the problems of limited prediction dimensions and subjective optimization in existing technologies.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the structure of the dental porcelain restoration color prediction system provided according to the embodiments of this application; Figure 2 This is a schematic diagram of a multimodal image acquisition module provided according to an embodiment of this application; Figure 3 This is a schematic diagram of an image recognition processing module according to an embodiment of this application; Figure 4 This is a schematic diagram of an intelligent color prediction module provided according to an embodiment of this application; Figure 5 This is a schematic diagram of a parameter output and optimization module provided according to an embodiment of this application; Figure 6 This is a schematic diagram of a color prediction system for dental porcelain restorations provided according to an embodiment of this application; Figure 7 This is a flowchart of a method for predicting the color of dental porcelain restorations according to an embodiment of this application; Figure 8 This is a schematic diagram of a method for predicting the color of dental porcelain restorations according to an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0018] 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 of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] The following describes an embodiment of the color prediction system for dental porcelain restorations according to the accompanying drawings. Addressing the issue of limited prediction dimensions mentioned in the background section, this application provides a color prediction system for dental porcelain restorations. In this system, a multimodal image acquisition module simultaneously acquires high-resolution color images of natural teeth, three-dimensional scan images of the prepared tooth surface, and microstructure images of the porcelain substrate, while simultaneously recording ambient light parameters. This overcomes the limitations of traditional single-acquisition methods, enabling full-dimensional data capture from the multi-faceted color gradation of natural teeth and tooth morphological features to the distribution of micropores in the porcelain layer. Combined with the image recognition processing module's noise reduction, CNN color feature extraction, and edge detection algorithms, saliva reflection and angular interference are eliminated, generating a natural tooth color space feature vector and tooth structure parameters, thus improving the comprehensiveness and accuracy of data acquisition. Furthermore, a ResNet and Tran... The Sformer hybrid architecture model integrates color features, morphological parameters, porcelain layer optical properties, and ambient light correction algorithms. This avoids the limitations of single-feature modeling and accurately simulates the light reflection and transmission effects of porcelain layers under different tooth morphologies. It reduces the color difference between the predicted color of the restoration and the natural tooth, improving the naturalness of color matching. Through parameter output and optimization modules, it automatically compares the predicted color with the natural tooth color, generating a quantitative deviation report and optimization schemes for layered porcelain layer thickness and firing temperature curves. This replaces traditional experience-based subjective adjustments, reducing porcelain powder loss and prolonged restoration cycles caused by repeated firing trials. Furthermore, AR visualization overlays the predicted effect onto the oral cavity scene, allowing doctors and patients to intuitively confirm the restoration effect, improving clinical diagnostic efficiency and ensuring a high degree of color matching between the restoration and the natural tooth. Therefore, it solves the problems of limited prediction dimensions and subjective optimization in existing technologies.

[0020] Figure 1 This is a schematic diagram of the structure of the color prediction system for dental porcelain restorations provided in the embodiments of this application.

[0021] This application provides a color prediction system for dental porcelain restorations, the system 10 including: Multimodal image acquisition module 100, image recognition and processing module 200, intelligent color prediction module 300, parameter output and optimization module 400.

[0022] The multimodal image acquisition module 100 is used to acquire high-definition color images of natural teeth in the oral cavity, three-dimensional scan images of tooth preparation surfaces, and microstructure images of the ceramic restoration substrate, while simultaneously recording the acquired ambient light parameters. The image recognition and processing module 200 is used to process the raw data and output the color space feature vector of natural teeth, the morphological parameters of tooth preparation surfaces, and the micropore distribution parameters of the ceramic substrate. The intelligent color prediction module 300 is used to construct a deep learning model that integrates image features and the optical properties of the ceramic layer. It takes color feature vectors, morphological parameters, micropore distribution parameters, and preset ceramic layer thickness parameters as input, and combines them with an ambient light correction algorithm to predict the final color data of the oral ceramic restoration. The parameter output and optimization module 400 is used to compare the predicted color data with the color features of natural teeth, generate a color deviation report, output suggestions for ceramic layer thickness adjustment and optimization schemes for firing process parameters, and support the visualization of the prediction effect.

[0023] It is understood that the embodiments of this application simultaneously acquire high-definition color images of natural teeth, three-dimensional scan images of the prepared tooth surface, and microstructure images of the porcelain substrate through a multimodal image acquisition module, while simultaneously recording ambient light parameters. This overcomes the limitations of traditional single acquisition, enabling full-dimensional data capture from the multi-faceted color gradation of natural teeth and tooth morphological features to the distribution of micropores in the porcelain layer. Combined with the image recognition processing module's noise reduction and correction, CNN color feature extraction, and edge detection algorithms, saliva reflection and angular interference are eliminated, generating natural tooth color space feature vectors and tooth structure parameters, thus improving the comprehensiveness and accuracy of data acquisition. Furthermore, a ResNet and Transformer hybrid architecture model is constructed through an intelligent color prediction module, integrating color... This approach utilizes color features, morphological parameters, porcelain layer optical properties, and ambient light correction algorithms to avoid the limitations of single-feature modeling. It accurately simulates the light reflection and transmission effects of the porcelain layer under different tooth morphologies, reducing the color difference between the predicted color of the restoration and the natural tooth, thus improving the naturalness of color matching. Through parameter output and optimization modules, it automatically compares the predicted color with the natural tooth color, generating a quantitative deviation report and optimization schemes for layered porcelain layer thickness and firing temperature curves. This replaces traditional subjective adjustments based on experience, reducing porcelain powder loss and prolonged restoration cycles caused by repeated firing trials. Furthermore, AR visualization overlays the predicted effect onto the oral cavity scene, allowing doctors and patients to intuitively confirm the restoration effect, improving clinical diagnostic efficiency and ensuring a high degree of color matching between the restoration and the natural tooth. Therefore, it solves the problems of limited prediction dimensions and subjective optimization in existing technologies.

[0024] In this embodiment of the application, the multimodal image acquisition module 100 further includes: Figure 2 As shown, the unit includes an intraoral high-definition imaging unit, a three-dimensional structural scanning unit, a substrate microscopic imaging unit, and an ambient light monitoring unit.

[0025] The intraoral high-definition imaging unit uses a macro camera with a color calibration card to acquire high-definition RGB images of the labial, lingual, and proximal surfaces of natural teeth; the three-dimensional structure scanning unit acquires three-dimensional point cloud data of the prepared tooth surface through structured light scanning; the substrate microscopic imaging unit uses microscopic imaging technology to acquire microscopic structural images of porcelain powder before sintering and identify particle size and pore distribution; and the ambient light monitoring unit is used to acquire color temperature and illuminance data in the oral cavity and treatment environment in real time.

[0026] Understandably, the high-definition intraoral imaging unit of this application uses a macro camera with a color calibration card to specifically acquire high-definition RGBB images of the labial, lingual, and proximal surfaces of natural teeth. The calibration card ensures the consistency of color data under different shooting scenarios, avoiding color distortion caused by saliva reflection and angular deviation. The three-dimensional structure scanning unit uses structured light scanning to acquire three-dimensional point cloud data of the prepared tooth surface, providing accurate spatial basis for subsequent extraction of morphological parameters such as area and curvature, and making up for the shortcomings of traditional prediction that ignore the influence of tooth morphology on the optical effect of the porcelain layer. The substrate microscopic imaging unit uses microscopic imaging technology to capture the microstructure of porcelain powder before sintering, identify particle size and porosity distribution, and enable the model to associate the microscopic characteristics and optical performance of the porcelain layer, breaking through the one-sidedness of traditional methods that only focus on macroscopic color. The ambient light monitoring unit collects color temperature and illuminance data of the treatment environment and the oral cavity in real time, providing input for the subsequent ambient light correction algorithm and eliminating the interference of different ambient lights on color judgment. The four units work together to ensure that subsequent image processing, color prediction, and process optimization have comprehensive and reliable data support, reducing the risk of restoration color mismatch due to missing or biased data from the source.

[0027] For example, a dental clinic performed a porcelain inlay restoration on a patient's maxillary lateral incisor. The patient's tooth was decayed and damaged, requiring removal of decayed material and preparation of the tooth surface. The restoration's color was required to harmonize naturally with the central incisor and canine on the same side. During the treatment, the multimodal image acquisition module was activated: the intraoral high-definition imaging unit used a 12-megapixel macro camera with a standard CIE color calibration card. The dentist adjusted the camera angle to avoid saliva reflection and acquired high-definition RGBB images of the labial surface (capturing the warm yellow gradation at the cervical region and the translucent gradation at the incisal edge), lingual surface (recording the base color of the exposed dentin area on the lingual side), and proximal surface (obtaining details of the color transition at the junction with the central incisor). The calibration card ensured the traceability of the acquired color data and avoided color deviations at different shooting times. The three-dimensional structure scanning unit used blue light structured light scanning technology to acquire three-dimensional images of the tooth preparation surface with an accuracy of 0.01mm. The point cloud data clearly presents the occlusal surface area (approximately 8 mm²), axial surface curvature (labial surface convexity 2.5 mm), and edge shoulder morphology (1 mm wide 90° right-angle shoulder) of the prepared surface. The substrate microscopic imaging unit uses 400x microscopic imaging technology to capture microscopic images of the selected glass-ceramic powder before sintering, automatically identifying particle size (average 3 μm) and pore distribution (porosity 2.1%) to determine whether the porcelain powder dispersion meets the optical uniformity requirements. The ambient light monitoring unit simultaneously collects data on natural light in the examination room (color temperature 5600 K, illuminance 380 lux) and local reflected light in the oral cavity (illuminance fluctuation ±50 lux due to mouth opening). All collected data are transmitted to subsequent modules in real time, laying a full-dimensional data foundation for accurately extracting color features and simulating the optical effects of the porcelain layer, avoiding the problem of missing key parameters in traditional single acquisition.

[0028] In this embodiment of the application, the image recognition processing module 200 includes: Figure 3 As shown, there are an image processing unit, a color feature extraction unit, and a structural parameter recognition unit.

[0029] The image processing unit is used to eliminate image interference caused by saliva reflection in the oral cavity and shooting angle through image denoising, white balance correction, and shadow removal algorithms; the color feature extraction unit is used to segment the natural tooth area and background through CNN image recognition algorithm, extract the CIELab value of each pixel, and generate a heat map of natural tooth color distribution and average color feature vector; the structural parameter recognition unit extracts the area, curvature and edge morphology parameters of the tooth preparation surface from the three-dimensional scan image through image segmentation and edge detection algorithms, and identifies the porosity and particle aggregation state parameters of the porcelain substrate from the microstructure image.

[0030] Understandably, the image processing unit of this application utilizes denoising, white balance correction, and shadow removal algorithms to specifically eliminate image interference caused by saliva reflection in the oral cavity and shooting angle deviations, avoiding distortion of the original data that could affect subsequent analysis. The color feature extraction unit uses a CNN image recognition algorithm to accurately segment the natural teeth and background areas, extracts CIELab values ​​pixel by pixel, and generates a color distribution heatmap and average feature vector, solving the problems of low efficiency and large errors in traditional manual color feature extraction, and ensuring accurate quantification of natural tooth color information. The structural parameter recognition unit extracts morphological parameters such as the area and curvature of the tooth preparation surface from the 3D scan image through image segmentation and edge detection, and identifies the porcelain porosity and particle aggregation state from the microscopic image, providing structured data for subsequent analysis, avoiding predictive bias caused by missing parameters, and finally outputting standardized feature data to provide high-quality input for the intelligent prediction module.

[0031] For example, a dental clinic performed a porcelain crown restoration on a patient's mandibular first molar. The multimodal image acquisition module had previously acquired multidimensional images of the tooth and adjacent teeth. However, the original images contained saliva reflection (the reflective area in the occlusal fissure accounted for about 15%), gingival shadows caused by the shooting angle (a 2mm wide dark area appeared at the mesial gingival margin), and slight blurring (caused by the patient's slight shaking). The image recognition and processing module then begins operation: the image processing unit first removes image blur and interference using a Gaussian filtering algorithm, then performs white balance correction to correct the bluish tint caused by the cold light in the examination room, and finally eliminates gingival shadows using a shadow removal algorithm to make the tooth outline and color transition natural; the color feature extraction unit calls a pre-trained CNN model to automatically segment the natural tooth area from the oral mucosa and tool background, and extracts CIELab values ​​pixel by pixel—the enamel area of ​​the occlusal surface has an L value (brightness) of approximately 85, an a value (red-green tint) of approximately 2.1, and a b value (yellow-blue tint) of approximately 18.5, and the dentin area at the cervical region has an L value of approximately 72, an a value of approximately 3.3, and a b value of approximately 22.8, generating a clear color distribution heatmap. The color gradient from the occlusal surface to the neck of the tooth is analyzed, and the average color feature vector of the natural tooth is calculated (L*=78.2, a*=2.7, b*=20.6). The structural parameter recognition unit extracts the area of ​​the tooth preparation surface (approximately 12 mm²), the buccal and lingual curvature (buccal curvature radius 8 mm), and the edge shoulder morphology (0.8 mm wide rounded shoulder) from the 3D scan image through an edge detection algorithm. Then, it identifies the porosity (1.8%) and particle aggregation state (no obvious aggregation, uniform particle dispersion) from the microscopic image of the porcelain powder. Finally, it outputs standardized color and structural parameters, providing interference-free and highly accurate input data for the intelligent color prediction module, avoiding the impact of defects in the original image on the accuracy of subsequent restoration color prediction.

[0032] In this embodiment of the application, the intelligent color prediction module 300 includes: Figure 4As shown, there are feature fusion unit, deep learning prediction model unit, and ambient light correction unit.

[0033] The feature fusion unit is used to nonlinearly fuse the color feature vectors and structural parameters obtained from image recognition with the optical parameters of porcelain layer thickness and refractive index to construct a multi-dimensional input feature matrix; the deep learning prediction model unit is based on a hybrid architecture of ResNet and Transformer, and is trained by color matching samples of natural teeth and restorations collected clinically, outputting the CIELab prediction value and color uniformity assessment of the restoration under the target ambient light; the ambient light correction unit is used to correct the predicted color data according to the ambient light monitoring data and the color temperature compensation algorithm to eliminate color deviation under different treatment environments.

[0034] Understandably, the model feature fusion unit in this application embodiment nonlinearly fuses color feature vectors, tooth structure parameters, porcelain layer thickness, and refractive index optical parameters to construct a multi-dimensional input feature matrix, enabling the model to comprehensively associate various factors affecting color. The deep learning prediction model unit is based on a hybrid architecture of ResNet and Transformer, trained using clinically matched samples. It retains ResNet's ability to extract local image features while leveraging Transformer to capture global correlations between parameters, accurately outputting the CIELab value and color uniformity assessment of the restoration under target ambient light, thus reducing prediction bias. The ambient light correction unit corrects the prediction results based on monitored color temperature and illuminance data using a color temperature compensation algorithm, eliminating interference from different treatment ambient lights on color presentation. This ensures that the predicted color matches the actual visual effect of the restoration after installation, regardless of whether it is under natural light in the clinic or indoor lighting, improving the reliability of clinical applications.

[0035] For example, a dental center performed an all-ceramic crown restoration on a patient's maxillary second premolar. Prior to this, the image recognition processing module had already output the tooth's color feature vector (L*=76.5, a*=2.9, b*=19.8), the tooth preparation surface structural parameters (area 10mm², buccal curvature radius 7mm, 0.9mm wide rounded shoulder), and simultaneously obtained the optical parameters of the selected zirconia ceramic layer (bottom layer thickness 0.3mm, middle layer thickness 0.5mm, surface layer thickness 0.2mm, refractive index 1.52) and the ambient light data (color temperature 6000K, illuminance 400lux). The intelligent color prediction module then begins operation: the feature fusion unit first nonlinearly fuses the color feature vector, structural parameters, porcelain thickness, and refractive index—using a weighted algorithm to correlate tooth curvature with the light reflection path of the porcelain layer, and combining porosity to correct for the influence of refractive index, constructing a 12-dimensional input feature matrix containing "color-morphology-optics"; the deep learning prediction model unit calls a model based on a hybrid ResNet and Transformer architecture (this model has been trained on 500 sets of natural tooth-restor matching clinical samples with an accuracy of 95%), and after inputting the feature matrix, outputs the CIELab predicted value of the restoration under the target ambient light (L*=76.3, (a*=3.0, b*=19.6), and simultaneously generated a color uniformity assessment report, showing a 92% uniformity in the translucency zone at the incisal edge and an 89% uniformity in color transition at the cervical region, with no obvious color gradation breaks; the ambient light correction unit, based on the monitored 6000K color temperature data, fine-tuned the predicted value through a color temperature compensation algorithm—correcting the slightly cool b* value from 19.6 to 19.7, eliminating the possibility of the restoration appearing whiter under cold light, and ultimately controlling the color difference between the predicted color and the natural tooth to within 1.2, ensuring that the restoration presents a natural appearance that coordinates with adjacent teeth in both the clinic environment and the patient's daily natural light environment, providing a precise color target basis for subsequent optimization of process parameters.

[0036] In this embodiment of the application, the parameter output and optimization module 400 includes, as follows: Figure 5 As shown, there are a color comparison unit, a process parameter generation unit, and a visualization display unit.

[0037] The color comparison unit is used to calculate the color difference between the predicted color and the natural tooth color; the process parameter generation unit is used to output the thickness adjustment parameters of the bottom, middle and top layers of the porcelain layer and the firing temperature curve optimization scheme based on the color deviation data; the visualization display unit is used to overlay the predicted restoration color onto the patient's oral scan image through AR, allowing doctors and patients to intuitively view the restoration effect.

[0038] Understandably, the color comparison unit in this application embodiment automatically calculates the color difference between the predicted color and the natural tooth color, generating a quantitative deviation report to replace traditional subjective visual assessment, making color differences more intuitive and traceable. Based on the deviation data, the process parameter generation unit accurately outputs the thickness adjustment parameters for the bottom, middle, and top layers of the porcelain layer, as well as the optimized firing temperature curve, avoiding repeated trial firings by doctors based on experience, reducing porcelain powder material waste, and shortening the restoration cycle. The visualization display unit uses AR technology to overlay the predicted restoration color onto the patient's oral cavity scan image, allowing doctors and patients to intuitively preview the restoration effect, improving doctor-patient communication efficiency, confirming color matching in advance, reducing the probability of secondary restoration, and promoting the standardized implementation of dental porcelain restorations.

[0039] For example, after a dental clinic performs color prediction for a patient's maxillary second premolar all-ceramic crown, the parameter output and optimization module immediately starts working: the color comparison unit first calls the original color feature vector of the natural tooth (L*=76.5, a*=2.9, b*=19.8) and the CIELab predicted value of the restoration output by the intelligent prediction module (L*=75.8, a*=3.1, b*=19.2), and calculates the color difference ΔE=1.3 using the CIEDE2000 formula (the clinically acceptable range ΔE<2, but there is still room for optimization); the process parameter generation unit automatically analyzes based on this deviation data—judging that the brightness is slightly low (L difference 0.7) and the red-green color is slightly high (a difference 0.2), and outputs a layer adjustment plan: increasing the thickness of the bottom layer of porcelain from 0.3mm to 0.32mm (to increase light reflectivity to increase brightness), and increasing the thickness of the middle layer of porcelain. Add 0.05mm and select a warm-toned porcelain powder (correcting the a* value), while keeping the surface layer at 0.2mm. At the same time, optimize the firing temperature curve: reduce the heating rate from 5℃ / min to 4℃ / min (to reduce the fluctuation of porcelain layer porosity), and extend the high-temperature holding time by 2min (to ensure color uniformity). The visualization unit then calls up the patient's oral cavity 3D scan image and uses AR technology to overlay the adjusted predicted restoration color onto the prepared tooth position. The doctor and patient can clearly see the color transition between the restoration and the first premolar and first molar on the same side—the enamel translucency of the occlusal surface and the dentin base color at the cervical region are highly coordinated with the adjacent teeth, without obvious color gradation breaks. After the patient intuitively confirms that the restoration effect meets expectations, the doctor directly sends the optimized process parameters to the porcelain processing workshop, avoiding repeated trial firings caused by traditional experience-based adjustments and significantly shortening the restoration cycle.

[0040] The color prediction system for dental porcelain restorations proposed in this application simultaneously acquires high-definition color images of natural teeth, three-dimensional scan images of the prepared tooth surface, and microstructure images of the porcelain substrate through a multimodal image acquisition module. It also records ambient light parameters, overcoming the limitations of traditional single-acquisition methods. This system captures data across all dimensions, from the multi-faceted color gradation of natural teeth and tooth morphology to the distribution of micropores in the porcelain layer. Combined with noise reduction, CNN color feature extraction, and edge detection algorithms in the image recognition processing module, it eliminates saliva reflection and angular interference, generating a natural tooth color space feature vector and tooth structure parameters, thus improving the comprehensiveness and accuracy of data acquisition. Furthermore, the system utilizes a hybrid ResNet and Transformer architecture built through an intelligent color prediction module. This novel approach integrates color features, morphological parameters, optical properties of the porcelain layer, and ambient light correction algorithms. It avoids the limitations of single-feature modeling and accurately simulates the light reflection and transmission effects of the porcelain layer under different tooth morphologies. This reduces the color difference between the predicted color of the restoration and the natural tooth, improving the naturalness of the color match. Through parameter output and optimization modules, it automatically compares the predicted color with the natural tooth color, generating a quantitative deviation report and optimization schemes for layered porcelain layer thickness and firing temperature curves. This replaces the subjective adjustments based on traditional experience, reducing porcelain powder loss and prolonged restoration cycles caused by repeated firing trials. Furthermore, AR visualization overlays the predicted effect onto the oral cavity scene, allowing doctors and patients to intuitively confirm the restoration effect, improving clinical diagnostic efficiency and ensuring a high degree of color matching between the restoration and the natural tooth. Therefore, it solves the problems of limited prediction dimensions and subjective optimization in existing technologies.

[0041] The following will illustrate the color prediction system for dental porcelain restorations through a specific embodiment, such as... Figure 6 As shown, it includes: A patient with long-term caries in his maxillary first molar, with a tooth loss area of ​​1 / 3, went to a dental specialist hospital for treatment. After evaluation, the doctor decided to customize a zirconia all-ceramic crown for him. It was necessary to use a dental porcelain restoration color prediction system to ensure that the color of the restoration was naturally coordinated with the adjacent teeth. The multimodal image acquisition module is activated first: the intraoral high-definition imaging unit uses a 16-megapixel macro camera equipped with a standard CIE1976 color calibration card. The dentist uses sterile cotton rolls to wipe the surface of the prepared teeth and adjacent teeth with saliva, adjusting the camera angle to 45° to avoid reflections in the occlusal fissures. High-definition RGBB images are acquired from the labial surface of the prepared teeth (capturing the translucency of the enamel and the warm yellow tone of the dentin at the neck), lingual surface (recording the color transition at the edge of the lingual tooth defect), and proximal surface (acquiring the color gradient at the junction with the second premolar and second molar). The calibration card simultaneously records the corners of the images to ensure the traceability of subsequent color data and avoid deviations caused by changes in lighting at different shooting times. The three-dimensional structure scanning unit activates the blue light structured light scanning function, scanning around the prepared tooth surface with a scanning accuracy of 0.005mm to generate three-dimensional point cloud data, clearly presenting the occlusion of the prepared surface. The surface area (approximately 14 mm²), buccal surface curvature radius (9 mm), lingual surface curvature radius (7.5 mm), and edge shoulder morphology (1 mm wide rounded shoulder, shoulder height 0.5 mm from the gingival margin); the substrate microscopic imaging unit uses a 400x metallurgical microscope to perform microscopic imaging on the selected 3Y-TZP zirconia porcelain powder (powder state before sintering), automatically identifying the particle size (average 2.8 μm) and pore distribution (porosity 1.5%) of the porcelain powder particles, confirming that the particles are uniformly dispersed without agglomeration, meeting the requirements of optical uniformity; the ambient light monitoring unit collects ambient light data in the examination room in real time—the current examination room uses 6500K cool white LED lights, with an illuminance stable at 380 lux, and simultaneously captures local reflected light fluctuations (±30 lux) caused by mouth opening in the oral cavity. These data and image data are synchronously stored in the system database to provide full-dimensional raw data for subsequent modules.

[0042] Upon receiving the multimodal acquisition data, the image recognition and processing module immediately addresses interference issues in the original image. The image processing unit first analyzes the intraoral image, identifying approximately 12% salivary reflection in the occlusal fissures, a 1.8mm wide shadow at the mesial gingival margin due to the shooting angle, and slight shaking during the patient's photography causing image edge blurring. To address this, the unit first smooths the blurred edges using a 5×5 Gaussian filter algorithm, filtering out unwanted pixels and restoring the clarity of the tooth surface texture. Next, it performs white balance correction using the grayscale world method, adjusting the blue channel brightness value based on the neutral grayscale pixels at the gingival-tooth junction to correct the overall bluish tint caused by the 6500K cold light, restoring the natural warm yellow tone of the tooth. Finally, using a region growing algorithm, the normal grayscale value at the tooth-gingival junction is used as the implantation point to gradually expand and fill the gingival shadow area, ensuring a smooth and natural transition between the tooth edge and the gingiva, eliminating light and dark discontinuities. After processing, the color feature extraction unit calls a pre-trained image segmentation model to automatically distinguish the natural tooth area from the background such as the oral mucosa and treatment tools. Then, it extracts the CIELab color values ​​of the natural tooth pixel by pixel: L value (brightness) of the enamel area on the occlusal surface is about 86.2, a value (red-green hue) is about 2.3, and b value (yellow-blue hue) is about 18.7; L value of the enamel-dentin mixed area in the middle of the tooth is about 79.5, a value is about 2.8, and b value is about 20.5; and L value of the dentin area at the cervical region is about 73.1, a value is about 3.2, and b value is about 22.3. Based on these data, a color distribution heatmap is generated, which clearly shows the color gradient from the occlusal surface to the cervical region. At the same time, the average color feature vector of the natural tooth is calculated (L=79.6, a*=2.8, b*=20.5). The structural parameter recognition unit uses the Canny edge detection algorithm to accurately extract key structural parameters of the tooth preparation surface from the 3D point cloud data: in addition to the area and curvature already obtained, it also identifies the occlusal surface depth (2.2mm) and axial convergence (6°) of the preparation surface to ensure that it meets the retention requirements of the all-ceramic crown; at the same time, it further verifies the porosity of 1.5% and the particle aggregation rate of less than 3% from the microscopic image of the porcelain powder to confirm the stability of the optical properties of the porcelain material. Finally, these standardized color features and structural parameters are packaged and transmitted to the intelligent color prediction module.

[0043] After receiving the input data, the intelligent color prediction module first integrates multi-dimensional parameters through the feature fusion unit. This unit nonlinearly fuses the average color feature vector of natural teeth (L*=79.6, a*=2.8, b*=20.5), the structural parameters of the prepared tooth surface (area 14mm², buccal curvature 9mm, etc.), and the preset optical parameters of zirconia porcelain layer (bottom layer thickness 0.3mm, middle layer thickness 0.5mm, surface layer thickness 0.2mm, refractive index 1.56). Through a weighted algorithm, the buccal curvature of the tooth is associated with the light reflection path of the porcelain layer (the greater the curvature, the smaller the deviation of the light reflection angle, and the weight is assigned 0.3). Combined with the porcelain powder porosity to correct the refractive index (for every 1% increase in porosity, the refractive index decreases by 0.01), a 15-dimensional input feature matrix containing the three dimensions of "color-morphology-optics" is finally constructed to ensure that the correlation between the parameters is fully captured. Subsequently, the deep learning prediction model unit was put into operation. The ResNet and Transformer hybrid architecture model on this unit has been trained and optimized through 800 sets of clinical natural tooth-restoration color matching samples (covering color data of different age groups and tooth positions), with a test set accuracy of 96.2%. After inputting a 15-dimensional feature matrix, the model combines the target ambient light data (6500K color temperature and 380 lux illuminance in the clinic) to output the CIELab prediction value of the restoration (L*=79.4, a*=2.9, b*=20.3), and generates a color uniformity assessment report: 93% uniformity of enamel translucency on the occlusal surface and 91% uniformity of color transition at the cervical region, with no obvious color gradation breaks, meeting clinical aesthetic requirements. Finally, the ambient light correction unit fine-tunes the predicted value based on the ambient light monitoring data using a color temperature compensation algorithm. Because the 6500K color temperature is too cool, it may cause the restoration to appear slightly white. The unit corrects the b value from 20.3 to 20.4, making the warm yellow tone of the restoration closer to natural teeth. At the same time, based on the fluctuation of local reflected light in the oral cavity, the L value is dynamically calibrated by ±0.1. Finally, the final predicted color of the restoration is determined to be (L*=79.5, a*=2.9, b*=20.4), with a color difference value ΔE=0.8 (calculated using the CIEDE2000 formula) from the average color of natural teeth, which is far below the clinically acceptable standard of ΔE<2.

[0044] The parameter output and optimization module then connects to the intelligent prediction results and performs subsequent processing: the color comparison unit first calls the average color feature vector of natural teeth and the final predicted color of the restoration, and calculates the color difference value ΔE=0.8 using the CIEDE2000 color difference formula, which is considered to be "excellent match". Although it meets clinical needs, in order to further improve aesthetics, the system still automatically analyzes the deviation details - the a value is slightly higher by 0.1 (slightly reddish) and the L value is basically the same, providing direction for subsequent optimization. Based on this deviation data and combined with the optical property database of the zirconia ceramic layer, the process parameter generation unit outputs a targeted optimization scheme: the thickness of the bottom layer remains unchanged at 0.3mm (to ensure strength), the middle layer uses A2 color grade ceramic powder with a warm tone, and the thickness is increased from 0.5mm to 0.52mm (to correct the a* value deviation by increasing the thickness of the warm tone ceramic layer), and the thickness of the surface transparent ceramic layer is maintained at 0.2mm (to ensure the transparency of the glaze); in terms of optimizing the firing temperature curve, the heating rate is reduced from 5℃ / min to 4.5℃ / min (to reduce the fluctuation of the internal porosity of the ceramic layer), the holding time in the high-temperature section (1350℃) is extended from 15min to 18min (to ensure uniform color sintering), and a stepped cooling is adopted in the cooling section (1350℃→1000℃, rate 3℃ / min; 1000℃→room temperature, rate 5℃ / min) to avoid color deviation caused by thermal stress. The visualization unit utilizes a 3D panoramic image of the patient's oral cavity. Using AR technology, the optimized predicted color of the restoration is precisely overlaid onto the prepared position of the maxillary first molar. Both doctor and patient can view the effect 360° on a high-definition display screen – the enamel texture on the occlusal surface of the restoration, and the color of the connection between the cervical region and the gingiva are highly coordinated with adjacent teeth, especially showing no significant color difference in the smile line area. After the patient confirms that the effect meets expectations, the doctor clicks the "Confirm Output" button in the system. The optimized porcelain layer thickness parameters (bottom layer 0.3mm, middle layer 0.52mm, top layer 0.2mm) and firing curve data are automatically and encryptedly transmitted to the cooperating porcelain processing workshop. Three days later, the completed all-ceramic crown arrives at the clinic. During the clinical trial, the measured color difference between the restoration and the natural tooth was 0.9, essentially consistent with the predicted value. The patient was highly satisfied with the restoration effect. The entire restoration cycle was shortened by 40% compared to the traditional experience-based trial firing method, and porcelain powder material consumption was reduced by 25%.

[0045] In summary, the embodiments of this application acquire multi-dimensional data on the color of natural teeth, tooth morphology, microscopic characteristics of porcelain layers, and ambient light, breaking through the limitations of traditional single-source acquisition. The image recognition processing module eliminates interference such as reflections and shadows, extracting accurate color features and structural parameters to provide a reliable basis for prediction. The intelligent color prediction module integrates multi-dimensional parameters and corrects for ambient light deviations, avoiding the one-sidedness of single-feature prediction. The parameter output and optimization module automatically compares and generates process plans and AR visualization previews, replacing subjective adjustments based on experience. Ultimately, the color difference between the restoration and the natural tooth is only 0.8-0.9, significantly improving color matching accuracy. At the same time, it shortens the restoration cycle by 40%, reduces porcelain powder waste by 25%, improves doctor-patient communication efficiency and patient satisfaction, and promotes the standardization and precision development of dental porcelain restoration.

[0046] In this embodiment, the method for constructing a color prediction system for dental ceramic restorations includes: configuring multimodal image acquisition hardware comprising an intraoral high-definition imaging unit, a three-dimensional structural scanning unit, a substrate microscopic imaging unit, and an ambient light monitoring unit; building an image recognition processing flow integrating denoising correction, CNN feature extraction, and edge detection algorithms; constructing an intelligent color prediction model based on a ResNet and Transformer hybrid architecture and incorporating an ambient light correction algorithm; designing a parameter output and optimization device comprising color comparison, layer thickness and firing process parameter generation, and AR visualization functions; and training the model with clinical samples and verifying the coherence of data flow and prediction accuracy of each device to complete the system construction.

[0047] It is understood that the embodiments of this application, by configuring multimodal image acquisition hardware, collect comprehensive data on natural teeth, tooth morphology, microscopic characteristics of porcelain layers, and ambient light, providing comprehensive raw data support for the system; establish an image recognition processing flow integrating multiple algorithms to effectively eliminate interference and extract accurate feature parameters, ensuring data quality; construct a prediction model with a fusion architecture and correction algorithm to avoid the one-sidedness of single feature prediction and improve color prediction accuracy; design a multifunctional parameter output and optimization device to replace subjective adjustments based on experience and provide intuitive preview; combine clinical sample training and data flow verification to ensure the reliable collaboration of all aspects of the system, ultimately building a standardized and accurate restoration color prediction system, solving the problems of rough acquisition and large prediction deviation in traditional restorations, and providing technical support for efficient and high-quality dental porcelain restorations in clinical practice.

[0048] For example, the dental medical equipment R&D team carried out the implementation work step by step to build a color prediction system for dental porcelain restorations: First, in the hardware configuration stage, the team purchased multimodal image acquisition hardware—the intraoral high-definition imaging unit selected a 16-megapixel macro camera (model MC-200) equipped with a standard CIE color calibration card, which is suitable for shooting in the narrow space of the oral cavity; the three-dimensional structure scanning unit adopted a blue light structured light scanner (model SL-800), with a scanning accuracy of 0.005mm, which meets the requirements for capturing details of tooth morphology; the substrate microscopic imaging unit was equipped with a 400x metallurgical microscope (model MX-500), which supports automatic identification of porcelain powder particle size and porosity; the ambient light monitoring unit selected a high-precision color temperature illuminance meter (model LS-100), which can simultaneously record the color temperature (3000K-10000K) and illuminance (0-1000lux) data of the treatment environment. All hardware achieved synchronous data transmission through industrial Ethernet, ensuring that the acquisition dimensions fully cover natural teeth, tooth morphology, porcelain characteristics and ambient light.

[0049] Next, the image recognition processing workflow was built: the team integrated multiple algorithm modules based on the Python + OpenCV framework—in the noise reduction and correction stage, the image was first blurred by a 5×5 Gaussian filter, then the white balance was corrected using the gray-world method, and finally the gingival shadow was eliminated by the region growing algorithm; in the CNN feature extraction stage, the lightweight MobileNetV2 network was used as the basic architecture, and 1000 labeled oral dental images (including natural teeth and background region labels) were used for training to achieve automatic segmentation of natural tooth regions, and then CIELab values ​​were extracted pixel by pixel to generate color features; in the edge detection stage, the Canny algorithm was integrated, and the edges of the tooth preparation surface in the 3D scan image were accurately extracted by dynamically adjusting the threshold (high threshold 80, low threshold 40), and finally an automated processing workflow of "original image input → interference elimination → feature extraction → parameter output" was formed, and the processing time for a single set of images was controlled within 30 seconds.

[0050] Subsequently, an intelligent color prediction model was constructed: the team built a hybrid architecture using ResNet50 (responsible for extracting local image features) and Transformer (responsible for capturing global parameter correlations). The 2048-dimensional features output by ResNet50 were spliced ​​with tooth structure parameters and porcelain optical parameters (refractive index 1.52-1.60, thickness 0.2-0.8mm) and then input into the Transformer module containing a 6-layer encoder. At the same time, an ambient light correction algorithm was incorporated to correct the prediction results based on the color temperature compensation formula. During the model training phase, 800 sets of clinical samples (covering different tooth positions of anterior and posterior teeth, including natural tooth color and actual restoration color data) were used and divided into training and testing sets in a 7:3 ratio. After 100 iterations using the Adam optimizer, the final color prediction accuracy of the testing set reached 96.5%, and the proportion of samples with a color difference value ΔE < 1.0 was 82%.

[0051] During the parameter output and optimization device design phase, the team developed a visualization software interface—the color comparison module integrates the CIEDE2000 formula to automatically calculate the color difference between the predicted color and the natural tooth color and generate a deviation report; the process parameter generation module has a built-in ceramic optical database and outputs layer thickness and firing temperature curves based on the deviation data; the AR visualization module interfaces with the Unity3D engine to overlay the predicted restoration color onto the patient's oral cavity 3D scan image, allowing doctors to rotate and zoom to view the effect via tablet gestures, while also reserving a data interface for interface with the ceramic processing workshop system.

[0052] Finally, clinical sample training and system validation were conducted: The team used 200 real restoration cases (covering different age groups and tooth positions) to incrementally train the model with newly collected clinical data and optimize parameter weights; at the same time, the continuity of data flow between various devices was verified—from multimodal hardware data acquisition (approximately 2 minutes / case), to image recognition processing (30 seconds / case), to intelligent prediction (10 seconds / case), and finally to output parameters and AR display (15 seconds / case), with no data gaps throughout the entire process; by comparing the predicted color with the actual restoration color, the measured average color difference value ΔE=0.92, which meets the clinical aesthetic requirements, and the entire system was finally built and can be put into clinical trial.

[0053] Next, with reference to the accompanying drawings, a method for predicting the color of dental porcelain restorations according to embodiments of this application is described.

[0054] like Figure 7 As shown, the method for predicting the color of dental porcelain restorations includes the following steps: In step S101, multimodal images of natural teeth in the oral cavity, microscopic images of ceramic restoration substrates, and ambient light parameters are acquired.

[0055] It is understood that the embodiments of this application comprehensively capture the color gradation and texture details of the labial, lingual, and proximal surfaces by acquiring multimodal images of natural teeth in the oral cavity, microscopic images of the ceramic restoration substrate, and ambient light parameters, avoiding the omission of key color information from a single angle; the microscopic images of the substrate can grasp the optical characteristics parameters of the ceramic powder particle size and pore distribution, providing a basis for subsequent simulation of the light reflection effect of the ceramic layer; the ambient light parameters can record the color temperature and illuminance fluctuations of the treatment scene, eliminating the influence of environmental interference on color judgment, and laying a solid foundation for accurate color matching between the restoration and the natural teeth.

[0056] In step S102, based on the multimodal images of natural teeth in the oral cavity, the microscopic images of the ceramic restoration substrate, and the ambient light parameters, the color feature vector of the natural teeth, the morphological parameters of the prepared tooth surface, and the microstructural parameters of the ceramic layer are extracted using an image recognition algorithm.

[0057] Image recognition algorithms refer to the methods by which computers process, analyze, and understand image information, extract target features, and achieve accurate recognition. The core is to transform visual information into quantifiable data or decision results. Typically, CNNs are first used to separate the target from the background, followed by edge detection, color analysis, and texture analysis to capture key parameters such as the target's shape, color, and texture, ultimately completing recognition tasks such as classification and matching.

[0058] It is understood that the embodiments of this application use image recognition algorithms to accurately process data such as multimodal images of natural teeth in the oral cavity, extract natural tooth color feature vectors, tooth preparation surface morphology parameters and porcelain layer microstructure parameters, transform fuzzy visual information into quantitative data, provide reliable input for intelligent color prediction, ensure the accuracy of subsequent restoration color prediction, and lay a solid data foundation for the accurate matching of restorations and natural teeth.

[0059] For example, taking the all-ceramic crown restoration of a patient's maxillary lateral incisor as an example, the system first acquires multimodal images of the tooth (including high-resolution images of the labial / lingual / proximal surfaces and a 3D scan of the prepared tooth surface), microscopic images of zirconia porcelain powder, and ambient light data of the examination room (color temperature 5500K, illuminance 350 lux). The image recognition algorithm then starts working: first, it uses a 5×5 Gaussian filter to eliminate the blurring of the image edges caused by the patient's slight shaking; then, it uses the gray-world method to correct the bluish tint caused by the cold light; and then it uses a region growing algorithm to remove the 1.5mm wide shadow at the gingival margin. Finally, it uses a lightweight CNN model to complete the target segmentation—from the oral cavity image... The process involves separating natural teeth from the gingiva and treatment tool background, locating the porcelain powder area from the microscopic image, and eliminating slide impurities. Subsequently, color analysis is used to extract the CIELab values ​​of natural tooth pixels, generating a color heatmap and feature vector (L*=77.2, a*=3.1, b*=20.3). An edge detection algorithm is used to extract morphological parameters such as the prepared surface area (8mm²) and buccal curvature radius (6mm) from the 3D image. Texture analysis is used to identify microscopic parameters such as porcelain powder porosity (1.6%) and particle size (3μm). The three types of quantitative parameters output provide accurate data support for subsequent color prediction. Taking the restoration of a patient's maxillary lateral incisor with an all-ceramic crown as an example, the system first acquires multimodal images of the tooth (including high-resolution images of the labial / lingual / proximal surfaces and a 3D scan of the prepared tooth surface), microscopic images of zirconia porcelain powder, and ambient light data of the examination room (color temperature 5500K, illuminance 350 lux). The image recognition algorithm then starts working: first, a 5×5 Gaussian filter is used to eliminate the blurring of the image edges caused by the patient's slight shaking; the gray-world method is used to correct the bluish tint caused by the cold light; then, a region growing algorithm is used to remove the 1.5mm wide shadow at the gingival margin; finally, a lightweight CNN model is used to complete the target segmentation—separating the tooth from the oral cavity image. The background of natural teeth, gingiva, and treatment tools was used to locate the porcelain powder area from the microscopic image and exclude glass slide impurities. Then, the CIELab values ​​of natural tooth pixels were extracted through color analysis to generate a color heatmap and feature vector (L*=77.2, a*=3.1, b*=20.3). Morphological parameters such as the prepared surface area (8mm²) and buccal curvature radius (6mm) were extracted from the 3D image through edge detection algorithm. Microscopic parameters such as porcelain powder porosity (1.6%) and particle size (3μm) were identified through texture analysis. The three types of quantitative parameters output at the end provide accurate data support for subsequent color prediction.

[0060] In step S103, the natural tooth color feature vector, tooth preparation surface morphology parameters, and porcelain layer microstructure parameters are input into the pre-trained intelligent color prediction model, and combined with the ambient light correction algorithm, the predicted color data of the restoration is obtained.

[0061] Among them, the intelligent color prediction model refers to a model that integrates the color characteristics of natural teeth, tooth morphology parameters, optical properties of porcelain layers, and ambient light data, based on a hybrid architecture of ResNet and Transformer, and combined with an ambient light correction algorithm, to accurately output the final color data of dental porcelain restorations.

[0062] It is understood that the embodiments of this application utilize an intelligent color prediction model to take into account the multi-dimensional parameters output by the image recognition algorithm, integrate the color features of natural teeth, tooth morphology, and microscopic characteristics of the porcelain layer, and combine them with an ambient light correction algorithm to eliminate interference, thus avoiding the one-sidedness of single parameter prediction. Based on the hybrid architecture of ResNet and Transformer, the model mines parameter correlations and accurately outputs the final color data of the restoration, providing a reliable basis for subsequent color comparison and process optimization, ensuring that the color of the restoration is highly consistent with the natural teeth, and improving the quality and efficiency of restoration.

[0063] In step S104, the predicted color data is compared with the natural tooth color to generate a color deviation report and optimized parameters for porcelain layer thickness and firing process, which are then displayed using AR.

[0064] Understandably, the embodiments of this application compare the predicted color data with the natural tooth color, which can accurately locate the color difference between the two, and the generated color deviation report provides a clear direction for process adjustment; at the same time, the output porcelain layer thickness and firing process optimization parameters can directly guide the porcelain processing stage to adjust the thickness of the bottom / middle / top porcelain layer and the firing temperature curve, avoiding material waste and cycle extension caused by traditional experience trial and error; and the AR display function can overlay the predicted restoration color onto the patient's oral scan image, allowing doctors and patients to intuitively perceive the restoration effect, improve the efficiency of doctor-patient communication, ensure that the restoration color is highly coordinated with the natural tooth, and guarantee the aesthetics and accuracy of clinical restoration.

[0065] For example, taking a patient's posterior teeth with large-area defects due to caries requiring all-ceramic inlay restoration: First, multimodal images of natural teeth in the oral cavity (including high-resolution images of the occlusal and proximal surfaces of posterior teeth), microscopic images of the ceramic inlay substrate, and ambient light parameters of the treatment room are collected. After processing by image recognition algorithms, the actual color data of the natural teeth on the same side as the patient's healthy posterior teeth are extracted (CIELab values: L*=75.3, a*=2.6, b*=19.4). At the same time, the intelligent color prediction model outputs the predicted color data of the inlay based on the characteristics of the ceramic material and the tooth morphology parameters (CIELab values: L*=75.3, a*=2.1, b*=17.2). Subsequently, the staff imported the two sets of CIELab data into the color comparison module. The module first calculated the overall color difference value using the CIEDE2000 color difference formula, obtaining ΔE=2.7 (exceeding the clinically acceptable range of ΔE<1.5). Then, it broke down the deviations dimension by dimension: a value deviation of 0.5 (predicted redness is lower than natural teeth, appearing whiter) and b value deviation of 2.2 (predicted yellowness is insufficient, appearing cooler). Then, a color deviation report was automatically generated. The report pages presented "Comparison Table of Original Data of Natural Teeth and Predicted Inlay Colors", "Interpretation of Deviation Values ​​and Meanings in Each Dimension", and "Overall Color Difference Level Assessment (marked 'Needs Adjustment')". At the same time, combined with the ceramic optical database, specific adjustment direction suggestions were attached at the end of the report - "Increase the thickness of the middle layer of yellow porcelain to improve the b value, and fine-tune the porcelain powder ratio to supplement the redness", providing a clear basis for subsequent optimization of process parameters. Finally, the adjusted predicted inlay color was superimposed on the defective tooth position in the patient's oral cavity 3D scan model through the AR function, allowing doctors and patients to intuitively view the color correction effect.

[0066] According to the color prediction method for dental porcelain restorations proposed in this application, this application embodiment simultaneously acquires high-definition color images of natural teeth, three-dimensional scan images of the prepared tooth surface, and microstructure images of the porcelain substrate through a multimodal image acquisition module, while simultaneously recording ambient light parameters. This overcomes the limitations of traditional single-acquisition methods, enabling full-dimensional data capture from the multi-faceted color gradation of natural teeth and tooth morphological features to the distribution of micropores in the porcelain layer. Combined with the image recognition processing module's noise reduction, CNN color feature extraction, and edge detection algorithms, saliva reflection and angular interference are eliminated, generating a natural tooth color space feature vector and tooth structure parameters, thus improving the comprehensiveness and accuracy of data acquisition. Furthermore, a ResNet and Transformer hybrid image prediction module is constructed through an intelligent color prediction module. This integrated architecture model combines color features, morphological parameters, porcelain layer optical properties, and ambient light correction algorithms to avoid the limitations of single-feature modeling. It accurately simulates the light reflection and transmission effects of the porcelain layer under different tooth morphologies, reducing the color difference between the predicted color of the restoration and the natural tooth, thus improving the naturalness of color matching. Through parameter output and optimization modules, it automatically compares the predicted color with the natural tooth color, generating a quantitative deviation report and optimization schemes for layered porcelain layer thickness and firing temperature curves. This replaces the subjective adjustments relying on traditional experience, reducing porcelain powder loss and prolonged restoration cycles caused by repeated firing trials. Furthermore, AR visualization overlays the predicted effect onto the oral cavity scene, allowing doctors and patients to intuitively confirm the restoration effect, improving clinical diagnostic efficiency and ensuring a high degree of color matching between the restoration and the natural tooth. Therefore, it solves the problems of limited prediction dimensions and subjective optimization in existing technologies.

[0067] The following will illustrate the method for predicting the color of dental porcelain restorations through a specific example, such as... Figure 8 As shown, it includes: The patient had long-term caries in their left maxillary central incisor, resulting in severe tooth structure loss. After diagnosis by the dentist, a full ceramic crown restoration was required. The entire restoration process was strictly carried out step by step according to the established method. First, data collection was carried out: the dentist used a high-definition intraoral endoscope to take images of the labial, lingual, and proximal surfaces of the patient's left maxillary central incisor (prepared tooth structure). At the same time, a 3D intraoral scanner was used to scan this area to obtain a 3D image containing details of the tooth preparation surface, forming a multimodal image of the natural tooth. Then, a small sample was taken from the zirconia ceramic block used for restoration (a commonly used dental restorative ceramic material) and placed under a metallurgical microscope to take a microscopic image of the ceramic material magnified 200 times, clearly showing the distribution of ceramic powder particles. In addition, the dentist used a professional color temperature illuminance meter to measure the ambient light parameters of the treatment area in the clinic. After multiple measurements and averaging, the color temperature was found to be 5500K and the illuminance was 380 lux, ensuring that subsequent color correlation analysis was not affected by ambient light fluctuations.

[0068] After data acquisition, the acquired images and parameters were processed. First, multimodal images of natural teeth and microscopic images of porcelain were imported into an image recognition tool. This tool first used a Gaussian filtering algorithm to eliminate bright spots caused by saliva reflection in the oral cavity images, and then used the grayscale world method to correct for lingual shadows on the teeth caused by the angle of the clinic's overhead lights, making the image details clearer. Next, using an image recognition algorithm, the natural tooth area was automatically segmented from the background areas such as the oral mucosa and treatment instruments. The CIELab color value of each pixel within the natural tooth area was accurately extracted, and the average value was calculated to obtain the natural tooth color feature vector, specifically (L*=76). (a*=2.9, b*=20.1); Simultaneously, the algorithm performs contour analysis on the 3D scan image to extract key morphological parameters of the tooth preparation surface, including the maximum cross-sectional area of ​​the preparation surface (8.2 mm²), the radius of curvature of the labial cervical margin curve (5.3 mm), and the occlusal gap of the occlusal surface (1.8 mm). For the microscopic image of the porcelain material, the algorithm obtains the microstructural parameters of the porcelain substrate through particle recognition and porosity statistics: the average particle size of the porcelain powder is 2.8 μm and the porosity is 1.5%. These parameters fully reflect the core characteristics of natural teeth and restorative substrates.

[0069] Next, the color prediction stage begins: the staff inputs the extracted natural tooth color feature vector (L*=76.8, a*=2.9, b*=20.1), tooth preparation surface morphology parameters (area 8.2mm², radius of curvature 5.3mm, etc.), and porcelain layer microstructure parameters (particle size 2.8μm, porosity 1.5%) into the pre-trained intelligent color prediction model. This model, based on a hybrid ResNet and Transformer architecture, has been trained and optimized using thousands of sets of dental restoration color data. It can accurately correlate tooth features, substrate properties, and the final restoration color. Simultaneously, the model incorporates previously collected ambient light parameters (color temperature 5500K, illuminance 380 lux) for ambient light correction. Since different ambient lights affect human color perception, the model uses algorithmic compensation to eliminate interference from ambient light on color prediction, ultimately outputting predicted color data for the all-ceramic crown. The corresponding CIELab values ​​are (L*=76.8, a*=2.8, b*=18.3). The prediction results also include a color stability assessment, showing that the color deviation rate under normal lighting conditions is less than 0.5%.

[0070] Finally, color comparison and plan confirmation were carried out: The doctor compared the predicted color data (L*=76.8, a*=2.8, b*=18.3) with the natural tooth color feature vector (L*=76.8, a*=2.9, b*=20.1). First, the CIEDE2000 color difference formula was used to calculate the overall color difference value ΔE=2.1, which exceeded the clinically acceptable range of ΔE<1.5. Then, a detailed color deviation report was generated, which included a "Comparison Table of Natural Tooth and Predicted Restoration Color Data", clearly listing each difference between the two sets of CIELab values, and interpreting the meaning of the deviation: a value deviation of 0.1 (the predicted redness is slightly lower than that of natural teeth, and it appears slightly pale), b value deviation of 1.8 (the predicted yellowness is insufficient, and the overall color is cold gray, which does not match the natural warm yellow tone of natural teeth), and marked the color difference level as "needs optimization and adjustment". Based on the deviation report and combined with the optical performance data of zirconia ceramic materials, optimized parameters for the ceramic layer thickness and firing process were automatically generated: the thickness of the warm yellow base ceramic layer in the all-ceramic crown was adjusted from the initial design of 0.2mm to 0.32mm, and the holding time at 1250℃ in the firing process was extended from 15 minutes to 18 minutes to enhance the warm color performance of the ceramic layer and improve color stability. Afterwards, the dentist used an AR display tool to overlay the adjusted predicted color of the restoration onto the left maxillary central incisor position of the patient's oral cavity 3D scan model in real time. The restoration effect was then shown to the patient via a tablet computer—the screen clearly showed that the restored all-ceramic crown was in harmony with the color of the healthy right central incisor, with consistent labial gloss. After visually confirming the effect, the patient expressed satisfaction. The dentist then generated a processing order based on the optimized parameters and sent it to the ceramic material processing center for the fabrication of the all-ceramic crown.

[0071] In summary, this application's embodiments lay the data foundation for subsequent analysis by collecting multimodal images of the patient's left maxillary central incisor, microscopic images of zirconia porcelain, and ambient light parameters of the treatment room. Image recognition algorithms are used to eliminate image interference, segment the target region, and extract the natural tooth color feature vector, tooth preparation surface morphology parameters, and porcelain layer microstructure parameters. These parameters are then input into a pre-trained intelligent color prediction model, combined with an ambient light correction algorithm, to output predicted color data for the restoration. A deviation report and process optimization parameters are generated through color comparison, and the effect is displayed using AR to coordinate the restoration color with the natural tooth color, meeting clinical restoration needs and patient expectations.

[0072] Figure 9 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0073] When the processor 902 executes the program, it implements the method for predicting the color of dental porcelain restorations provided in the above embodiments.

[0074] Furthermore, electronic devices also include: Communication interface 903 is used for communication between memory 901 and processor 902.

[0075] The memory 901 is used to store computer programs that can run on the processor 902.

[0076] The memory 901 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0077] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0078] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0079] The processor 902 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0080] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting the color of dental porcelain restorations.

[0081] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-described method for predicting the color of dental porcelain restorations.

[0082] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0083] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0084] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0085] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0086] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0087] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A color prediction system for dental porcelain restorations, characterized in that, include: The system includes a multimodal image acquisition module, an image recognition and processing module, an intelligent color prediction module, and a parameter output and optimization module; among which, The multimodal image acquisition module is used to acquire high-definition color images of natural teeth in the oral cavity, three-dimensional scan images of tooth preparation surfaces, and microstructure images of ceramic restoration substrates, while simultaneously recording the ambient light parameters. The image recognition processing module is used to process the raw data and output the natural tooth color space feature vector, tooth preparation surface morphology parameters, and ceramic substrate micropore distribution parameters. The intelligent color prediction module is used to construct a deep learning model that integrates image features and the optical properties of the ceramic layer. It takes the color feature vector, morphological parameters, micropore distribution parameters and preset ceramic layer thickness parameters as input, and combines them with an ambient light correction algorithm to predict the final color data of the dental ceramic restoration. The parameter output and optimization module is used to compare the predicted color data with the color characteristics of natural teeth, generate a color deviation report, output suggestions for adjusting the porcelain layer thickness and optimization schemes for firing process parameters, and support the visualization of the prediction effect.

2. The color prediction system for dental porcelain restorations according to claim 1, characterized in that, The multimodal image acquisition module includes an intraoral high-definition imaging unit, a three-dimensional structure scanning unit, a substrate microscopic imaging unit, and an ambient light monitoring unit. The intraoral high-definition imaging unit uses a macro camera with a color calibration card to acquire high-definition RGBB images of the labial, lingual, and proximal surfaces of natural teeth. The three-dimensional structure scanning unit acquires three-dimensional point cloud data of the prepared tooth surface through structured light scanning. The substrate microscopic imaging unit uses microscopic imaging technology to acquire microscopic structural images of porcelain powder before sintering and identify particle size and porosity distribution. The ambient light monitoring unit is used to acquire color temperature and illuminance data in the intraoral cavity and treatment environment in real time.

3. The color prediction system for dental porcelain restorations according to claim 1, characterized in that, The image recognition processing module includes an image processing unit, a color feature extraction unit, and a structural parameter recognition unit. The image processing unit eliminates image interference caused by saliva reflection and shooting angle through image denoising, white balance correction, and shadow removal algorithms. The color feature extraction unit segments the natural tooth region from the background using a CNN image recognition algorithm, extracts the CIELab value of each pixel, and generates a natural tooth color distribution heatmap and average color feature vector. The structural parameter recognition unit extracts the area, curvature, and edge morphology parameters of the tooth preparation surface from the 3D scan image using image segmentation and edge detection algorithms, and identifies the porosity and particle aggregation state parameters of the porcelain substrate from the microstructure image.

4. The color prediction system for dental porcelain restorations according to claim 1, characterized in that, The intelligent color prediction module includes a feature fusion unit, a deep learning prediction model unit, and an ambient light correction unit. The feature fusion unit nonlinearly fuses color feature vectors and structural parameters obtained from image recognition with optical parameters such as porcelain layer thickness and refractive index to construct a multi-dimensional input feature matrix. The deep learning prediction model unit, based on a ResNet and Transformer hybrid architecture, is trained using clinically collected natural teeth and restoration color matching samples, outputting the CIELab predicted value and color uniformity assessment of the restoration under target ambient light. The ambient light correction unit corrects the predicted color data based on ambient light monitoring data using a color temperature compensation algorithm, eliminating color deviations under different treatment environments.

5. The color prediction system for dental porcelain restorations according to claim 1, characterized in that, The parameter output and optimization module includes a color comparison unit, a process parameter generation unit, and a visualization unit. The color comparison unit is used to calculate the color difference between the predicted color and the natural tooth color. The process parameter generation unit is used to output thickness adjustment parameters for the bottom, middle, and top layers of the porcelain layer, as well as an optimization scheme for the firing temperature curve, based on the color deviation data. The visualization unit is used to overlay the predicted restoration color onto the patient's oral scan image using AR, allowing doctors and patients to intuitively view the restoration effect.

6. A method for constructing a color prediction system for dental porcelain restorations according to any one of claims 1-5, characterized in that, include: The system is equipped with multimodal image acquisition hardware, including an intraoral high-definition imaging unit, a three-dimensional structural scanning unit, a substrate microscopic imaging unit, and an ambient light monitoring unit. Establish an image recognition and processing workflow that integrates denoising correction, CNN feature extraction, and edge detection algorithms; Construct an intelligent color prediction model based on a hybrid architecture of ResNet and Transformer and incorporating ambient light correction algorithms; The design includes a parameter output and optimization device with functions such as color comparison, layer thickness and firing process parameter generation, and AR visualization. The system was constructed by training the model with clinical samples and verifying the consistency of data flow and prediction accuracy of each device.

7. A method for predicting the color of dental porcelain restorations, characterized in that, include: Acquire multimodal images of natural teeth in the oral cavity, microscopic images of ceramic restoration substrates, and ambient light parameters; Based on the multimodal images of natural teeth in the oral cavity, the microscopic images of the ceramic restoration substrate, and the ambient light parameters, the color feature vector of natural teeth, the morphological parameters of the tooth preparation surface, and the microstructural parameters of the ceramic layer are extracted using image recognition algorithms. The natural tooth color feature vector, tooth preparation surface morphology parameters, and porcelain layer microstructure parameters are input into a pre-trained intelligent color prediction model, and combined with an ambient light correction algorithm, to obtain the predicted color data of the restoration. The predicted color data is compared with the natural tooth color to generate a color deviation report and optimized parameters for porcelain layer thickness and firing process, which are then displayed using AR.

8. An electronic device, characterized in that, include: The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method for predicting the color of dental porcelain restorations as described in claim 7.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they implement the method for predicting the color of dental porcelain restorations as described in claim 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed, they implement the method for predicting the color of dental porcelain restorations as described in claim 7.

Images