Method for converting a view of a 3D model of a dental arch into a photo-realistic view

Abstract

A method for converting a "raw" view of a "raw" digital three-dimensional model into a hyper-realistic view, the method comprising the following steps: 21) creating a "conversion" learning library consisting of more than 1000 "conversion" records, each conversion record including: - a "conversion" photo representing the scene, and - a view or "conversion view" of a "conversion" digital three-dimensional model modeling the scene, the conversion view representing the scene as the conversion photo; 22) training at least one "conversion" neural network through the conversion learning library; 23) submitting the raw view to the at least one conversion neural network such that the at least one conversion neural network converts the raw view into a hyper-realistic view.

Description

Technical Field

[0001] The present invention relates to the field of analysis of dental arch photographs.

[0002] It particularly relates to a method for making three-dimensional models and views of such models hyper-realistic, creating a learning library designed to train a neural network based on these hyper-realistic views, and using the neural network thus trained to analyze photographs of dental arches. Background Art

[0003] The state of the art uses neural networks based on images (usually X-rays) to evaluate dental conditions, especially for forensic identification.

[0004] "Neural network" or "artificial neural network" is a set of algorithms well known to those skilled in the art. In particular, the neural network can be selected from:

[0005] - A neural network dedicated to image classification (referred to as a convolutional neural network (CNN)), such as:

[0006] - AlexNet (2012)

[0007] - ZF Net (2013)

[0008] - VGG Net (2014)

[0009] - GoogleNet (2015)

[0010] - Microsoft ResNet (2015)

[0011] - Caffe: BAIR Reference CaffeNet, BAIR AlexNet

[0012] - Torch: VGG_CNN_S, VGG_CNN_M, VGG_CNN_M_2048, VGG_CNN_M_1024, VGG_CNN_M_128, VGG_CNN_F, 16-layer VGG ILSVRC-2014, 19-layer VGG ILSVRC-2014, Network in Network (Imagenet and CIFAR-10)

[0013] - Google: Inception (V3, V4);

[0014] - Networks dedicated to locating and detecting objects in images (object detection networks), such as:

[0015] - R-CNN (2013)

[0016] - SSD (Single Shot MultiBox Detector: object detection network), Faster R-CNN (Faster Region-based Convolutional Neural Network: object detection network)

[0017] - Faster R-CNN (2015)

[0018] - SSD (2015).

[0019] The above list is not exhaustive.

[0020] For it to be operational, a neural network must be trained using a learning process called "deep learning" based on either an unpaired learning library or a paired learning library.

[0021] A paired learning library consists of a set of records, each record including an image and a description of that image. By presenting the records at the input of the neural network, the neural network gradually learns how to generate a description for the images presented to it.

[0022] For example, each record in the learning library can include an image of a dental arch and a representation identifying teeth or "tooth regions" in that image and a description of the corresponding tooth numbers. After being trained, the neural network will thus be able to recognize the representation of teeth and the corresponding tooth numbers in the images presented to it.

[0023] The quality of the analysis performed by the neural network depends directly on the number of records in the learning library. This learning library typically contains more than 10,000 records.

[0024] In the dental field, it is difficult to create a large number of records due to the limited number of images produced, especially by orthodontists and dentists, and due to the general confidentiality of these images.

[0025] The quality of the analysis performed by the neural network also depends on the quality of the descriptions of the records in the learning library. These descriptions are typically produced by an operator who delimits the tooth regions by computer and, after identifying the corresponding tooth (e.g., "right upper canine"), assigns a number to it accordingly. This operation is called tagging. If the operator makes a mistake when identifying or entering a tooth, the description will be incorrect, resulting in a decrease in the training quality.

[0026] The operator performing the tagging may have different interpretations of the same image. Thus, the quality of the learning library will depend on the interpretation adopted by the operator.

[0027] Therefore, there is a continuing need for a method for creating a high-quality learning library.

[0028] An object of the present invention is to meet this need. Summary of the Invention

[0029] The present invention proposes a method for enriching a historical learning library, the method comprising the following steps:

[0030] 1) Generating a digital three-dimensional model, or "historical model", of a scene, in particular a dental scene, and preferably a description of the historical model;

[0031] 2) Creating a hyper-realistic or "photo-realistic" view of the historical model;

[0032] 3) Preferably, based on the description of the historical model, creating a description of the hyper-realistic view or "historical description";

[0033] 4) Creating a historical record consisting of the hyper-realistic view and the description of the hyper-realistic view, and adding the historical record to the historical learning library.

[0034] As will be seen in more detail in the remainder of the specification, the enrichment method according to the invention uses models, in particular scans performed by dental professionals, to create hyper-realistic views equivalent to photographs. Thus, the present invention advantageously makes it possible to generate a learning library that makes it possible to train a neural network to analyze photographs, even if the learning library does not necessarily contain photographs.

[0035] Preferably, in step 1), a description of the historical model is generated, and in step 3), the historical description is created at least in part from the description of the historical model.

[0036] Preferably, the historical model is divided into a plurality of basic models, and thus in step 1), a specific description for each basic model to be represented in the basic model, preferably the hyper-realistic view, is generated in the description of the historical model, and in step 3), the specific description of the representation of the basic model in the hyper-realistic view is included in the historical description, at least a part of the specific description being inherited from the specific description.

[0037] For example, a basic model or "tooth model" representing a tooth is created in the historical model, and in the description of the historical model, a specific description is created for each tooth model, for example in order to identify the corresponding tooth number. It is thus very easy to fill in the historical description accordingly. In particular, the tooth numbers of the tooth models can be assigned to the representations of these tooth models in the hyper-realistic view. Advantageously, once the historical model and its description have been created, a computer can be used to generate the historical record without manual intervention. Thus, the creation of the historical description can be at least partially automated. This advantageously limits the risk of errors.

[0038] Furthermore, the enrichment method according to the invention advantageously makes it possible to generate a large number of historical records by modifying the views of the same model. Thus, the enrichment method preferably comprises the following steps after step 4):

[0039] 5) Modify the hyper-realistic view and then return to step 3).

[0040] In a preferred embodiment, the enrichment method comprises the following step 6) after step 4) or the optional step 5):

[0041] 6) Deform the historical model and then return to step 1).

[0042] Step 6) is particularly advantageous. Specifically, this makes it possible to create various historical models that are not obtained solely from measurements of the patient, in particular scans of the patient's dental arch. In particular, historical models can be created to simulate dental conditions for which few photos are available, such as conditions related to rare diseases.

[0043] Thus, the invention also relates to a method for analyzing an "analysis" photo representing the dental arch of a "patient" for "analysis", the method comprising the following steps:

[0044] A) Create a historical learning library containing more than 1000 historical records by implementing the enrichment method according to the invention;

[0045] B) Train at least one "analysis" neural network with the historical learning library;

[0046] C) Submit the analysis photo to the trained analysis neural network to obtain a description of the analysis photo.

[0047] When the historical learning library contains historical records related to a specific pathology, the analysis neural network thus advantageously makes it possible to evaluate whether the dental scene represented in the analysis photo corresponds to this pathology.

[0048] The invention also relates to a method for converting an "original" view of an "original" digital three-dimensional model, in particular a dental arch model, into a hyper-realistic view, the method comprising the following steps:

[0049] 21) Create a "conversion" learning library containing more than 1000 "conversion" records, each conversion record comprising:

[0050] - A "conversion" photo representing the scene, and

[0051] - A view or "conversion view" of the "conversion" digital three-dimensional model modeling the scene, the conversion view representing the scene in the same way as the conversion photo;

[0052] 22) Train at least one "transformation" neural network through a transformation learning library;

[0053] 23) Submit the original view to the at least one trained transformation neural network so that it determines the hyper-realistic view.

[0054] As will be seen in more detail in the rest of the specification, the transformation method is based on a neural network that is trained to be able to make the views of the model hyper-realistic. Thus, using this method advantageously makes it possible to create a hyper-realistic view library, thereby providing information that is substantially the same as a photograph without having to take a photo.

[0055] The transformation method can be particularly used to create hyper-realistic views of a historical model from the original views of the historical model in order to enrich the historical learning library according to the enrichment method of the present invention.

[0056] Preferably, in step 23), before submitting the original view to the transformation neural network, it is processed by a 3D engine. The result obtained thereby is further improved.

[0057] In one embodiment, the method includes the following additional steps:

[0058] 24) Associate the hyper-realistic view with a historical description to form a historical record of the historical learning library, that is, to perform steps 1) to 4).

[0059] The present invention also relates to a texturing method for making a "raw" digital three-dimensional model hyper-realistic, the method including the following steps:

[0060] 21') Create a "texture" learning library consisting of more than 1000 "texture" records, each texture record including:

[0061] - A non-realistic texture model representing a scene (especially the dental arch) and a description of the model, the description indicating that the model is a non-realistic texture, or

[0062] - A realistic texture model representing a scene (especially the dental arch) and a description of the model, the description indicating that the model is a realistic texture;

[0063] 22') Train at least one "texture" neural network through the texture learning library;

[0064] 23') Submit the original model to the at least one trained texture neural network so that it textures the original model to make it hyper-realistic.

[0065] As will be seen in more detail in the rest of the specification, this method advantageously makes it possible to create hyper-realistic views by simply observing the original model that has become hyper-realistic.

[0066] To this end, the method further comprises the following steps:

[0067] 24') Obtaining a hyper-realistic view by observing the original model that has become hyper-realistic in step 23').

[0068] The method according to the invention is at least partially, preferably fully implemented by a computer. Accordingly, the invention also relates to:

[0069] - A computer program comprising program code instructions for performing one or more steps of any method according to the invention when the program is executed by a computer,

[0070] - A storage medium on which such a program is stored, such as a memory or a CD-ROM.

[0071] Definition

[0072] "Patient" means anyone on whom the method according to the invention can be performed, whether or not the person is undergoing orthodontic treatment.

[0073] "Dental care professional" is understood to mean anyone qualified to provide dental care, particularly including orthodontists and dentists.

[0074] "Tooth condition" defines a set of characteristics related to a patient's dental arch at a given moment, such as the position of the teeth, the shape of the teeth, the position of orthodontic devices, etc. at that moment.

[0075] "Model" is understood to mean a digital three-dimensional model. It consists of a set of voxels. A "dental arch model" is a model representing at least a part of the dental arch and preferably at least 2, preferably at least 3, and preferably at least 4 teeth.

[0076] For the sake of clarity, a distinction is made between "dividing" a model into "multiple basic models" and "segmenting" an image (especially a photograph) into "multiple basic regions". The basic models and basic regions are 3D or 2D representations of elements of the real scene (such as teeth), respectively.

[0077] A model observed under defined viewing conditions, particularly from a defined angle and at a defined distance, is called a "view".

[0078] "Image" is a two-dimensional representation of a scene (formed by pixels). Thus, a "photo" is a specific image taken using a camera, usually a color image. "Camera" should be understood to mean any device that allows taking pictures, including video cameras, mobile phones, tablets or computers. A view is another example of an image.

[0079] A tooth property is a property whose value is specific to a tooth. Preferably, the value of the tooth property is assigned to each tooth area in the view under discussion or to each tooth model of the dental arch model under discussion. In particular, the tooth property is not related to the view or the entire model. It obtains its value from the characteristics of the tooth to which it relates.

[0080] A "scene" consists of a set of elements that can be observed simultaneously. A "tooth scene" is a scene that includes at least a part of the dental arch.

[0081] A "photo of the dental arch", a "representation of the dental arch", a "scan of the dental arch", a "model of the dental arch", or a "view of the dental arch" is understood to mean a photo, a representation, a scan, a model, or a view of all or part of the said dental arch.

[0082] The "acquisition conditions" of a photo or a view specify the spatial position and orientation of the device (camera) used to acquire the photo or the device used to acquire the view with respect to the dental arch of the patient (actual acquisition conditions), or the spatial position and orientation with respect to the model of the dental arch of the patient (virtual acquisition conditions). The acquisition conditions preferably also specify the calibration of the acquisition device. When the acquisition conditions correspond to a simulation in which the acquisition device is in the said acquisition conditions (theoretical positioning and preferably calibration of the acquisition device) with respect to the model, the acquisition conditions are called "virtual".

[0083] Under the virtual acquisition conditions of a view, the acquisition device can also be called "virtual". The view is specifically acquired by a conceptual acquisition device that has the characteristics of a "real" camera and whose characteristics have been used to acquire a photo that can be superimposed on the view.

[0084] The "calibration" of the acquisition device consists of all the values of the calibration parameters. A "calibration parameter" is a parameter inherent to the acquisition device (different from its position and its orientation) whose value affects the photo or the acquired view. The calibration parameters are preferably selected from the group consisting of the aperture diameter, the exposure time, the focal length, and the sensitivity.

[0085] "Discriminatory information" is characteristic information ("image features") that can generally be extracted from the image by computer processing of the image.

[0086] The discriminatory information can have a variable number of values. For example, depending on whether a pixel belongs to the contour, the contour information can be equal to 1 or 0. The brightness information can take a large number of values. Image processing makes it possible to extract and quantify the discriminatory information.

[0087] The discriminatory information can be represented in the form of a "chart". Thus, a chart is the result of processing an image to display the discriminatory information (such as the contours of the teeth and the gums).

[0088] The "degree of match" or "goodness of fit" between two objects is the name given to measure the difference between these two objects. If the degree of match is obtained by an optimization that minimizes the said difference, then the degree of match is the maximum ("best fit").

[0089] The photograph and the view showing the maximum degree of match represent the scene in substantially the same way. In particular, in the tooth scene, the representation of the teeth in the photograph and the view can substantially overlap.

[0090] The search for the view showing the maximum degree of match with the photograph is performed by searching for the virtual acquisition conditions of the view that show the maximum degree of match with the actual acquisition conditions of the photograph.

[0091] The comparison between the photograph and the view is preferably obtained by the comparison of two corresponding graphs. "Distance" is the name conventionally given to measure the difference between two graphs or between a photograph and a view.

[0092] A "learning library" is a database recorded by a computer suitable for training a neural network.

[0093] The training of the neural network is suitable for the desired goal and does not pose any particular difficulty to those skilled in the art.

[0094] Training the neural network consists in making it correspond to a learning library that contains information about two types of objects that the neural network must learn to "match" (i.e., be interconnected).

[0095] Training can be carried out according to a "paired" learning library, which consists of "paired" records, that is to say, each record includes a first object of the first type for the input of the neural network and a corresponding second object of the second type for the output of the neural network. It can also be said that the input and output of the neural network are "paired". Training the neural network with all these paired records will teach it to provide the corresponding object of the second type from any object of the first type.

[0096] For example, in order to enable a conversion neural network to convert an original view into a hyper-realistic view, it is trained with a conversion learning library so that when a corresponding converted view is presented to it at the input, it substantially provides a converted photograph at the output. In other words, all conversion records are provided to the conversion neural network, that is to say, each paired record contains a converted view (view of a dental arch model (first object of the first type)) and a corresponding converted photograph (photograph of the same dental arch, which is observed in the same way as the dental arch model to obtain the view (second object of the second type)), so that the conversion neural network determines its parameter values, so that when a converted view is presented to the conversion neural network at the input, the conversion neural network converts the converted view into a hyper-realistic view substantially the same as the corresponding photograph (if it has been taken).

[0097] Figure 12 An exemplary conversion record is shown.

[0098] Generally, such training is performed by providing a conversion view at the input to the conversion neural network and a converted photo at the output.

[0099] Similarly, the analysis neural network is trained by providing a history to the analysis neural network to learn from an analysis library so that the analysis neural network determines its parameter values such that when presenting a hyper-realistic view at the input to the analysis neural network, it provides a description that is substantially the same as the historical description corresponding to the hyper-realistic view.

[0100] Generally, such training is performed by providing a hyper-realistic view at the input to the analysis neural network and a historical description at the output.

[0101] The article “Image-to-Image Translation with Conditional Adversarial Networks” by Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A. Efros of the Berkeley AI Research (BAIR) Lab at the University of California, Berkeley shows the use of a paired learning library.

[0102] Training based on a paired learning library is preferred.

[0103] As an alternative, training can be performed based on a learning library referred to as “unpaired” or “unmatched”. Such a learning library consists of:

[0104] - an “output” set consisting of first objects of a first type, and

[0105] - an input set consisting of second objects of a second type,

[0106] The second objects do not necessarily correspond to the first objects, that is, they are independent of the first objects.

[0107] The input set and the output set are provided at the input and output of the neural network to train it. Such training of the neural network teaches it to provide the corresponding objects of the second type from any object of the first type.

[0108] For example, this "unpaired" training technique is described in the article "Unpaired image-to-image translation using cycle-consistent adversarial networks" by Zhu, Jun-Yan, et al.

[0109] Figure 13 A view of a 3D model of an input set of an unpaired learning library and an example of a photograph of an output set of the learning library are shown. The photograph does not correspond to the view, particularly because the dental arches are not observed in the same way and / or because the observed dental arches are the same.

[0110] For example, the input set can include a plurality of non-photorealistic texture models (a plurality of first objects) each representing a dental arch, and the output set can include a plurality of photorealistic texture models (a plurality of second objects) each representing a dental arch. Even if the dental arches represented in the input set are different from the dental arches represented in the output set, the "unpaired" training technique allows the neural network to learn to determine a corresponding second type of object (texture model) for a first type of object (non-texture model).

[0111] Of course, the quality of the learning depends on the number of records in the input set and the output set. The number of records in the input set is preferably substantially the same as the number of records in the output set.

[0112] According to the present invention, the unpaired learning library preferably includes a plurality of input sets and a plurality of output sets, each input set and each output set respectively including more than 1000, more than 5000, preferably more than 10000, preferably more than 30000, preferably more than 50000, and preferably the first 100000 first objects and second objects.

[0113] The nature of the object is not exhaustive. The object can be, for example, an image of the object or a set of information or "description" about the object. The description contains the values of the attributes of another object. For example, the attributes of an image of a dental scene can be used to identify the number of the represented tooth. Then the attribute is "tooth number", and for each tooth, the value of the attribute is the number of the tooth.

[0114] In this specification, for clarity, the qualifiers "historical", "original", "transformed", and "analyzed" are used.

[0115] Unless otherwise specified, "comprising" or "including" or "exhibiting" must be interpreted as non-restrictive. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] Other features and advantages of the present invention will become more apparent by reading the following detailed description and examining the drawings, wherein:

[0117] - Figures 1 to 3 and Figure 11 schematically illustrate the respective steps of the enrichment, analysis, transformation, and texturing methods according to the present invention;

[0118] - Figure 4 shows an example of an arch model;

[0119] - Figure 5 shows Figure 4 an example of the original view of the model in;

[0120] - Figure 6 shows Figure 5 an example of a hyper-realistic view obtained from the original view of;

[0121] - Figure 7 shows an exemplary transformed photograph of the dental arch;

[0122] - Figure 8 shows Figure 7 an exemplary transformed view corresponding to the transformed photograph in;

[0123] - Figure 9 shows an example of a transformed chart related to the tooth profile obtained from the transformed photograph;

[0124] - Figure 10a and Figure 10b respectively show the original view and the corresponding hyper-realistic view of the historical model, and the description of the hyper-realistic view inherits certain descriptions of the historical model;

[0125] - Figure 12 shows a record in the "paired" learning library, where the left-hand image shows the transformed view of the arch model to be provided at the input of the neural network, and the right-hand image shows the corresponding transformed photograph to be provided at the output of the neural network;

[0126] - Figure 13 shows an example of the view of the model in the input set of the unpaired learning library and the photograph in the output set of the learning library. Detailed Description of the Invention

[0127] The following detailed description is a detailed description of the preferred embodiments, but not limiting.

[0128] Create a historical learning library

[0129] The method for enriching a historical learning library according to the present invention includes steps 1) to 3).

[0130] In step 1), a historical model of the dental arch of a so-called "historical" patient is generated.

[0131] A historical model can be prepared based on measurements performed on the teeth of a historical patient or on a dental cast of his teeth (e.g., a plaster cast).

[0132] The historical model is preferably obtained from the real situation and preferably created using a 3D scanner. Such a model, called a "3D" model, can be observed from any angle.

[0133] In one embodiment, the historical model is theoretical, that is, it does not correspond to the real situation. In particular, a historical model can be created by combining a set of tooth models selected from a digital library. The arrangement of the tooth models is defined such that the historical model is realistic, that is, it corresponds to a situation that a patient may encounter. In particular, the tooth models are arranged in an arc according to their nature and are realistically oriented. Using a theoretical historical model advantageously makes it possible to simulate dental arches showing rare features.

[0134] Preferably, a description of the historical model is also generated.

[0135] The "description" of the model consists of a set of data related to the whole of the model or to a part of the model (e.g., the part of the model that models the teeth).

[0136] The historical model is preferably divided. In particular, for each tooth, a model of the tooth or a "tooth model" is preferably defined based on the historical model.

[0137] In the historical model, the tooth model is preferably delimited by the gingival margin, which can be decomposed into an inner gingival margin (relative to the tooth towards the inside of the mouth), an outer gingival margin (relative to the tooth oriented towards the outside of the mouth), and two lateral gingival margins.

[0138] One or more tooth attributes are associated with the tooth model based on the tooth modeled by the tooth model.

[0139] The tooth attributes are preferably attributes that are only related to the tooth modeled by the tooth model.

[0140] The tooth attributes are preferably selected from tooth number, tooth type, shape parameters of the tooth (e.g., tooth width (especially mesiopalatal width), thickness, crown height, mesial and distal deflection indices of the incisal edge, or level of wear), appearance parameters of the tooth (especially indicators of the presence of tartar, dental plaque or food on the tooth, translucency index or color parameters), or parameters related to the condition of the tooth (e.g., "worn", "broken", "decayed" or "fitted" (i.e., in contact with a dental device (e.g., an orthodontic device))), or parameters related to pathologies associated with the tooth (e.g., parameters related to the presence of gingivitis, MIH (molar incisor hypomineralization), AIH (autoimmune hepatitis), fluorosis or necrosis in the tooth area).

[0141] Tooth property values can be assigned to each tooth property of a specific tooth model.

[0142] For example, the tooth property "tooth type" will have the values "incisor", "canine", or "molar" respectively according to whether the tooth model is a model of an incisor, a canine, or a molar.

[0143] The tooth property "pathological condition" will have the values "healthy tooth", "broken tooth", "chipped tooth", "cracked tooth", "restored tooth", "tattooed tooth", or "carious tooth".

[0144] The assignment of tooth property values to the tooth model can be manual or at least partially automatic.

[0145] Similarly, tooth numbers are usually assigned according to standard rules. Therefore, it is sufficient to know the rule and the number of the tooth modeled by the tooth model to calculate the numbers of other tooth models.

[0146] In a preferred embodiment, the shape of a specific tooth model is analyzed in order to define its tooth property values, such as its number. This shape recognition can be performed manually. Preferably, it is performed by a neural network.

[0147] The definition of the tooth model and the tooth property values associated therewith form part of the description of the historical model.

[0148] Similarly, basic models other than tooth models can be defined based on the historical model, in particular models for the tongue, and / or the mouth, and / or the lips, and / or the jaw, and / or the gums, and / or dental devices (preferably orthodontic devices), and property values for the tongue, and / or the mouth, and / or the lips, and / or the jaw, and / or the gums, and / or the dental devices are assigned to them respectively.

[0149] The tongue property can be related to, for example, the position of the tongue (e.g., taking the value "retracted").

[0150] The mouth property can be related to, for example, the opening of the patient's mouth (e.g., taking the values "mouth open" or "mouth closed").

[0151] The orthodontic device property can be related to, for example, the presence of the dental device and / or the condition of the dental device (e.g., taking the values "device intact", "device damaged", or "device broken").

[0152] The description of the historical model can also include data related to the entire model, i.e., the value for "model property".

[0153] For example, the model attributes can define whether the dental condition shown by the historical model is "pathological" or "non-pathological" without performing an examination of each tooth. The model attributes preferably define one or more pathologies suffered by the historical patient when creating the historical model.

[0154] For example, the model attributes can also define the occlusion class, the position of the mandible relative to the maxilla ("overbite" or "overjet"), the overall hygiene index or the congestion index.

[0155] Convert to a hyper-realistic view

[0156] In step 2), a hyper-realistic view of the historical model is created, i.e., a view that looks like a photograph.

[0157] Preferably, the "original" view of the historical model is selected and then made hyper-realistic. The original view is preferably an extraoral view, for example, a view corresponding to a photograph preferably taken facing the patient using a retractor.

[0158] Any method of making the original view hyper-realistic is feasible. Preferably, a so-called "transformation" neural network is used, which is trained to make the original view hyper-realistic and includes steps 21) to 23).

[0159] Image conversion techniques are described in the article "Unpaired image-to-image translation using cycle-consistent adversarial networks" by Zhu, Jun-Yan et al. However, the article does not describe the conversion of the views of the model.

[0160] In step 21), a so-called "transformation" learning library consisting of more than 1000 so-called "transformation" records is created, and each transformation record includes:

[0161] - a "transformation" photograph representing a dental scene, and

[0162] - a view or "transformation view" of a "transformation" digital three-dimensional model that models the scene, and the transformation view represents the scene in the same way as the transformation photograph.

[0163] When the representation of the scene in the transformation view is substantially the same as that in the transformation photograph, the transformation view represents the scene in the same way as the transformation photograph.

[0164] The conversion learning library preferably contains more than 5000, preferably more than 10000, preferably more than 30000, preferably more than 50000, and preferably more than 100000 conversion records. The greater the number of conversion records, the better the ability of the conversion neural network to convert the original view into a hyper-realistic view.

[0165] For "converted" patients, conversion records are preferably generated as follows:

[0166] 211) Generate a model of the dental arch of the converted patient or a "conversion model";

[0167] 212) Under actual acquisition conditions, preferably obtain a conversion photo representing the dental arch by means of a mobile phone;

[0168] 213) Search for suitable virtual acquisition conditions to obtain a "converted" view of the conversion model that exhibits the maximum degree of match with the conversion photo under the virtual acquisition conditions, and obtain the converted view;

[0169] 214) Combine the conversion photo and the converted view to form a conversion record.

[0170] In particular, step 213) can be performed as described in WO 2016 / 066651.

[0171] Preferably, the conversion photo is processed to generate at least one "conversion" chart that at least partially represents discrimination information. Thus, the conversion chart represents discrimination information in the reference system of the conversion photo.

[0172] The discrimination information is preferably selected from the group consisting of contour information, color information, density information, distance information, brightness information, saturation information, information about reflection, and combinations of these information.

[0173] Those skilled in the art know how to process the conversion photo to reveal discrimination information.

[0174] For example, Figure 9 is a conversion chart related to the contour of the teeth obtained from the Figure 7 conversion photo.

[0175] Then the search includes the following steps:

[0176] i) Determine the virtual acquisition conditions to be tested;

[0177] ii) Generate a reference view of the conversion model under the virtual acquisition conditions to be tested;

[0178] iii) Process the reference view to generate at least one reference chart that at least partially represents discrimination information;

[0179] iv) Compare the transformation chart and the reference chart in order to determine the value of an evaluation function, the value of which depends on the difference between the transformation chart and the reference chart and corresponds to the decision to continue or stop the search for the following virtual acquisition conditions, which are closer to the actual acquisition conditions more precisely than the virtual acquisition conditions to be tested and determined in the instance of the previous step i);

[0180] v) If the value of the evaluation function corresponds to the decision to continue the search, modify the virtual acquisition conditions to be tested and then return to step ii).

[0181] Step i) involves starting from determining the virtual acquisition conditions to be tested, which are the virtual positions and orientations that may correspond to the actual position and orientation of the camera when capturing the transformation photo, but may preferably also be the virtual calibration that may correspond to the actual calibration of the camera when capturing the transformation photo.

[0182] In step ii), the camera is then virtually configured under the virtual acquisition conditions to be tested in order to obtain a reference view of the transformation model under these virtual acquisition conditions to be tested. Thus, the reference view corresponds to a photo of the transformation model, optionally calibrated, of the transformation model taken by the camera (if it were placed) under the virtual acquisition conditions to be tested.

[0183] In step iii), the reference view is processed in the same way as the transformation photo in order to generate a reference chart representing the discrimination information from the reference view.

[0184] In step iv), in order to compare the transformation photo and the reference view, their respective discrimination information is compared on the transformation chart and the reference chart. The difference or "distance" between these two charts is particularly evaluated by a score. For example, if the discrimination information is the contour of a tooth, the average distance between the tooth contour points appearing on the reference chart and the corresponding contour points appearing on the transformation chart can be compared, and the higher the score, the smaller the distance.

[0185] The score can be, for example, a correlation coefficient.

[0186] The score is then evaluated using an evaluation function. The evaluation function makes it possible to decide whether the loop from step i) to step v) should be continued or stopped. In step v), if the value of the evaluation function indicates the decision to continue the loop, the virtual acquisition conditions to be tested are modified and the loop in steps i) to v) is restarted, which includes generating the reference view and the reference chart, comparing the reference chart with the transformation icon to determine the score, and then making a decision based on that score.

[0187] Modify the virtual acquisition conditions to be tested corresponding to virtual movement in space and / or modify the orientation of the camera and / or preferably modify the calibration of the camera. The modification is preferably guided by heuristic rules, for example, guided by a favorable modification that seems most favorable for increasing the score based on the analysis of the previously obtained scores.

[0188] Continue the loop until the value of the evaluation function indicates a decision to stop the loop (e.g., if the score reaches or exceeds a threshold).

[0189] The virtual acquisition conditions are preferably optimized using metaheuristic methods, preferably evolutionary methods (preferably simulated annealing algorithm). Such methods are well known for non-linear optimization.

[0190] Preferably, it is selected from the group consisting of:

[0191] - Evolutionary algorithms, which are preferably selected from:

[0192] Evolution strategies, genetic algorithms, differential evolution algorithms, estimation of distribution algorithms, artificial immune systems, hybrid complex evolutionary path recombination, simulated annealing algorithms, ant colony algorithms, particle swarm optimization algorithms, tabu search, and GRASP methods;

[0193] - Kangaroo algorithms,

[0194] - Fletcher and Powell methods,

[0195] - Noise methods,

[0196] - Stochastic tunneling methods,

[0197] - Random restart hill climbing,

[0198] - Cross-entropy methods, and

[0199] - Hybrid methods between the above metaheuristic methods.

[0200] If the loop is exited without being able to obtain a satisfactory score, for example, if the score fails to reach the threshold, the method can stop (failure case) or restart with new discrimination information. The method can also continue with the virtual acquisition conditions corresponding to the best obtained score.

[0201] If the loop has been exited in the case of a satisfactory score that has been able to be obtained, for example, because the score reaches or even exceeds the threshold, the virtual acquisition conditions basically correspond to the actual acquisition conditions of the converted photo, and the reference view has the maximum degree of match with the converted photo. The representations of the tooth scenes in the reference view and the converted photo can basically overlap.

[0202] Then, select a reference view representing the tooth scene in the same manner as the converted photo as the converted view.

[0203] In step 22), the conversion neural network is trained by the conversion learning library. Such training is well known to those skilled in the art.

[0204] Generally, the training includes providing all the converted views at the input of the conversion neural network and providing all the converted photos at the output of the conversion neural network.

[0205] Through this training, the conversion neural network will learn how to convert any view of the model into a hyper-realistic view.

[0206] In step 23), the original view of the historical model is submitted to the conversion neural network. The conversion neural network converts the original view into a hyper-realistic view.

[0207] As an alternative to steps 21) to 23), step 2) may include the following steps: first, make the historical model hyper-realistic, and then extract the hyper-realistic views therefrom.

[0208] 21') Create a "texture" learning library consisting of more than 1000 "texture" records, each texture record including:

[0209] - A non-realistic texture model representing the dental arch (e.g., a scan of the dental arch) and a description of the model indicating that the model is a non-realistic texture, or

[0210] - A realistic texture model representing the dental arch (e.g., a scan of the dental arch made hyper-realistic) and a description of the model indicating that the model is a realistic texture, or

[0211] 22') Train at least one "texture" neural network by the texture learning library;

[0212] 23') Submit the historical model to the trained at least one texture neural network so that it textures the historical model to make it hyper-realistic.

[0213] Then, the hyper-realistic views can be directly obtained by observing the hyper-realistic historical model.

[0214] "Texturing" is understood to mean converting the model to give it a hyper-realistic appearance similar to what an observer of a real dental arch might observe. In other words, an observer of the hyper-realistic texture model has the impression of observing the dental arch itself.

[0215] In step 21'), the non-realistic texture model can be generated as described above to generate the historical model.

[0216] A photorealistic texture model can be generated by texturing an initial non-photorealistic texture model. Preferably, a method for generating a hyper-realistic model is implemented, the method including steps A") to C"), where the original model is the initial non-photorealistic texture model.

[0217] In step 22'), in particular, the training can be carried out according to the teachings in the article by Zhu, Jun-Yan et al. in "Unpaired image-to-image translation using cycle-consistent adversarial networks" (Open access Computer Vision Foundation).

[0218] Through this training, the texture neural network learns to texture the model so that it becomes hyper-realistic. In particular, it learns to texture the dental arch model.

[0219] In step 2), a hyper-realistic view of the 3D model can also be obtained by processing the original image with the aid of a conventional 3D engine.

[0220] A 3D engine is a software component that enables the simulation of the effects of the environment on the corresponding actual object on a digital three-dimensional object, in particular the lighting effects, optical effects, physical effects, and mechanical effects on the corresponding actual object. In other words, the 3D engine simulates the physical phenomena at the origin of these effects in the real world on a digital three-dimensional object.

[0221] For example, based on the relative position of the "virtual" light source with respect to the digital three-dimensional object and the nature of the light projected by the light source, the 3D engine will calculate the appearance of the object, for example, to display shadows or reflections. Thus, when the corresponding actual object is illuminated in the same way as the digital three-dimensional object, the appearance of the digital three-dimensional object will simulate the appearance of the actual object.

[0222] The 3D engine is also referred to as a 3D rendering engine, a graphics engine, a game engine, a physics engine, or a 3D modeler. In particular, such an engine can be selected from the following engines or their variants:

[0223] - Arnold

[0224] - Aqsis

[0225] - Arion renderer

[0226] - Artlantis

[0227] - Atomontage

[0228] - Blender

[0229] - Brazil r / s

[0230] - BusyRay

[0231] - Cycles

[0232] - FinalRender

[0233] - Fryrender

[0234] - Guerilla Renderer

[0235] - Indigo

[0236] - Iray

[0237] - Kerkythea

[0238] - KeyShot

[0239] - Kray

[0240] - Lightscape (Render Master)

[0241] - LightWorks

[0242] - Lumiscaphe

[0243] - LuxRender

[0244] - Maxwell Renderer

[0245] - Mental Ray

[0246] - Nova

[0247] - Octane

[0248] - Povray

[0249] - RenderMan

[0250] - Redsdk, Redway3d

[0251] - Sunflow

[0252] - Turtle

[0253] - V-Ray

[0254] - VIRTUALIGHT

[0255] - YafaRay。

[0256] In a particularly advantageous embodiment, the original view is first processed by the 3D engine and then submitted to the conversion neural network, as described above (step 23). Combining these two techniques can achieve significant effects.

[0257] In one embodiment, the original view can first be submitted to the conversion neural network and then processed by the 3D engine. However, this embodiment is not preferred.

[0258] In one embodiment, the 3D engine processes a hyper-realistic view directly obtained by observing the textured hyper-realistic historical model according to steps 21') to 23'). This additional processing also improves the realistic appearance of the obtained image.

[0259] In step 3), a description of the hyper-realistic view is created.

[0260] The description of the hyper-realistic view includes a set of data related to the whole of the view or a part of the view (e.g., related to the part of the view representing the teeth).

[0261] In the same way as the description of the historical model, the description of the hyper-realistic view can include the values of the properties of the teeth, and / or the tongue, and / or the mouth, and / or the lips, and / or the jaw, and / or the gums and / or the dental device represented in the hyper-realistic view. The above properties used for the description of the historical model can be the properties of the description of the hyper-realistic view.

[0262] The description of the hyper-realistic view can also include the values of the view properties, that is, the values of the view properties related to the hyper-realistic view or the whole original view. The view properties can particularly relate to:

[0263] - the position and / or orientation and / or calibration of the virtual camera used to obtain the original view, and / or

[0264] - the quality of the hyper-realistic view, in particular the quality of the hyper-realistic view related to the brightness, contrast or sharpness of the hyper-realistic view, and / or

[0265] - the content of the original view or the hyper-realistic view, such as the content related to the arrangement of the represented objects (e.g., to specify that the tongue covers certain teeth), or the content related to the therapeutic or non-therapeutic situation of the patient.

[0266] The description of the hyper-realistic view can be formed at least partially manually.

[0267] Preferably, the description is preferably generated at least partially, preferably completely, by a computer program by inheriting the historical model.

[0268] In particular, if the historical model has been partitioned, the virtual acquisition conditions make it possible to determine the basic models of the historical model represented in the hyper-realistic view and their respective positions. The values of the attributes related to the basic models available in the description of the historical model can thus be assigned to the same attributes related to the representation of the basic models in the hyper-realistic view.

[0269] For example, if the historical model has been partitioned to define a dental model and the description of the historical model specifies the number of the dental model, the same number can be assigned to the representation of the dental model in the hyper-realistic view.

[0270] Figure 10a An original view of the historical model, which has been partitioned to define a dental model, is shown. The description of the historical model contains the dental number of the historical model.

[0271] Thus, the values of at least some of the attributes of the description of the hyper-realistic view can inherit from the description of the historical model.

[0272] In step 4), a historical record consisting of the hyper-realistic view and the description of the hyper-realistic view is created and added to the historical learning library.

[0273] The historical learning library can consist only of historical records generated according to the enrichment method of the present invention. As an alternative, the historical learning library can contain historical records generated according to the enrichment method of the present invention and other historical records, such as other historical records created according to conventional methods, in particular by tagging photos.

[0274] In step 5) (which is optional), the hyper-realistic view of the historical model is modified and then the process returns to step 3).

[0275] To modify the hyper-realistic view, preferably a new hyper-realistic view is created from a new original view.

[0276] Thus, by executing the loop of steps 3) to 5), a large number of historical records corresponding to various observation conditions of the historical model can be created. Thus, a single historical model makes it possible to create many historical records without even the need for photos.

[0277] In step 6), preferably the historical model is deformed.

[0278] The deformation can particularly include:

[0279] - Moving the dental model, for example to simulate the spacing between two teeth,

[0280] - Deforming the dental model, for example to simulate bruxism,

[0281] - Deleting the dental model,

[0282] - Deform the jaw model.

[0283] In one embodiment, the deformation simulates pathology.

[0284] Step 6) obtains a theoretical historical model, which advantageously enables easy simulation of tooth conditions where measurements are unavailable.

[0285] Then return to step 2). Thus, based on the initial historical model, a historical record related to a tooth condition different from the tooth condition corresponding to the initial historical model can be obtained. In particular, historical records can be created for historical models corresponding to different stages of rare diseases.

[0286] The historical learning library preferably contains more than 5000, preferably more than 10000, preferably more than 30000, preferably more than 50000, and preferably more than 100000 historical records.

[0287] Analysis of the analyzed photo

[0288] To analyze the analysis photo, perform steps A) to C).

[0289] The method preferably includes a preparatory step in which a camera is used to obtain the analysis photo, and the camera is preferably selected from a mobile phone including a photo acquisition system, a so-called "connected" camera, a so-called "smartwatch", a tablet computer, or a fixed or portable personal computer. The camera is preferably a mobile phone.

[0290] More preferably, when obtaining the analysis photo, the distance between the camera and the dental arch is greater than 5 cm, greater than 8 cm, or even greater than 10 cm, which prevents water vapor from condensing on the optics of the camera and facilitates focusing. In addition, preferably, the camera, especially the mobile phone, does not have any specific optics for obtaining the analysis photo, which is possible especially due to the distance from the dental arch during acquisition.

[0291] The analysis photo is preferably a color photo, preferably a true color photo.

[0292] The analysis photo is preferably obtained by the patient, preferably without using a retainer to fix the camera, and especially without using a tripod.

[0293] In step A), a historical learning library containing historical records obtained by the enrichment method according to the present invention is created.

[0294] In step B), the "analysis" neural network is trained by the historical learning library. Such training is well known to those skilled in the art.

[0295] The neural network can be selected, in particular, from the list provided in the foregoing part of the present specification.

[0296] Through this training, the analysis neural network learns to evaluate the values of the attributes evaluated in the historical description for the photos presented to it.

[0297] For example, each historical description can specify a value ("yes" or "no") for the attribute "Is there malocclusion?".

[0298] The training generally includes providing all the said hyper-realistic views as inputs to the analysis neural network and providing all the said historical descriptions at the output of the analysis neural network.

[0299] In step C), the analysis photo is presented to the analysis neural network, and thus an evaluation of various attributes (such as "yes") is obtained, where for the case of the presence of malocclusion, the probability is 95%.

[0300] This analysis method can be used for therapeutic or non-therapeutic purposes, such as for research purposes or for purely aesthetic purposes.

[0301] For example, this analysis method can be used to evaluate the teeth of a patient during orthodontic treatment or teeth whitening treatment. It can be used to monitor the movement of teeth or the evolution of dental pathologies.

[0302] In one embodiment, the patient takes the analysis photo and implements the method, for example, with his mobile phone and a computer integrated into the mobile phone, the mobile phone being capable of communicating with the computer. Thus, by transmitting one or preferably multiple photos of the patient's teeth, the patient can very easily request an analysis of his dental condition without even having to move.

[0303] Analyzing the analysis photo is particularly useful for detecting rare diseases.

[0304] Simulate the dental situation

[0305] The conversion method according to the present invention can also be implemented to generate hyper-realistic views representing the dental condition simulated by a digital three-dimensional model of the dental arch. In particular, the dental condition can be simulated at a past or future simulation time in the context of a therapeutic or non-therapeutic process.

[0306] Therefore, the present invention relates to a method for simulating a dental condition, comprising the following steps:

[0307] A') At an update time, generating a digital three-dimensional model of the patient's dental arch, called the "update model", preferably as described above in step 1);

[0308] B') Deform the updated model so as to simulate the effect of the time between the update time and a simulation time before or after (e.g., more than 1 week, 1 month or 6 months before or after) the update time, to obtain a "simulation model", preferably as described in step 6) above;

[0309] C') Obtain a view of the simulation model or "original simulation view";

[0310] D') Convert the original simulation view into a hyper-realistic simulation view according to the conversion method of the present invention.

[0311] Thus, the hyper-realistic simulation view looks the same as a photograph taken at the simulation time. The hyper-realistic simulation view can be presented to the patient so as to, for example, present to the patient his future or past dental situation, thereby motivating him to undergo orthodontic treatment.

[0312] In step A'), preferably, as described above in step 1), the updated model is preferably divided into a plurality of basic models. Thus, in step B'), the deformation can be obtained, for example, by the movement or deformation of one or more basic models, in particular one or more tooth models, so as to simulate the effect of the orthodontic device.

[0313] Conversion of the model

[0314] The view of the original model that becomes hyper-realistic according to the conversion method of the present invention can be advantageously used to make the original model itself hyper-realistic.

[0315] Thus, the present invention also relates to a method for generating a hyper-realistic model from an original model, in particular from an original model of an dental arch, the method comprising the following consecutive steps:

[0316] A”) Obtain an original view of the original model;

[0317] B”) Convert the original view into a hyper-realistic view according to the conversion method of the present invention;

[0318] C”) For each pixel of the hyper-realistic view, identify the corresponding voxel of the original model, that is to say the voxel represented by the pixel in the hyper-realistic view, and assign the attribute value of the pixel to the attribute of the voxel.

[0319] The attribute of the pixel can particularly relate to its appearance, for example to its color or its brightness. The attribute of the voxel is preferably the same as the attribute of the pixel. Thus, for example, the color of the pixel is assigned to the voxel.

[0320] The method according to the present invention is at least partially, preferably fully implemented by a computer. Any computer can be considered, in particular a PC, a server or a tablet.

[0321] In particular, a computer typically includes: a processor, a memory, a human-machine interface (the human-machine interface typically includes a screen), and a module for communicating via the Internet, via Wi-Fi, via or via a telephone network. The software configured to implement the method of the present invention under discussion is loaded into the memory of the computer.

[0322] The computer can also be connected to a printer.

[0323] Of course, the present invention is not limited to the above-described and illustrated embodiments.

[0324] In particular, the patient is not limited to humans. The method according to the present invention can be used for another animal.

[0325] The learning library does not necessarily consist of "paired" records. It can be unpaired records.

[0326] The conversion learning library can include, for example:

[0327] - An input set consisting of "input views", each input view representing a view of a converted digital three-dimensional model for modeling a dental scene, preferably more than 1,000, more than 5,000, preferably more than 10,000, preferably more than 30,000, preferably more than 50,000, and preferably more than 100,000 input views, and

[0328] - An "output" set consisting of "output photos", each output photo representing a dental scene, and the output photos are preferably more than 1,000, more than 5,000, preferably more than 10,000, preferably more than 30,000, preferably more than 50,000, preferably more than 100,000,

[0329] The output photos can be independent of the output views, that is, they do not represent the same dental scene.

[0330] The texture learning library can include, for example:

[0331] - An input set consisting of non-photorealistic texture models, each non-photorealistic texture model representing an arch, and the non-photorealistic texture models are preferably more than 1,000, more than 5,000, preferably more than 10,000, preferably more than 30,000, preferably more than 50,000, preferably more than 100,000, and

[0332] - An output set consisting of photorealistic texture models, each photorealistic texture model representing an arch, and the photorealistic texture models are preferably more than 1,000, more than 5,000, preferably more than 10,000, preferably more than 30,000, preferably more than 50,000, and preferably more than 100,000,

[0333] The realistic texture model can be independent of the non-realistic texture model, that is, it does not represent the same dental arch.

Claims

1. A method for converting an original view of an original digital three-dimensional model of a dental arch into a hyper-realistic view, the method comprising the following steps: 21) Create a conversion learning library that is unpaired or consists of more than 1000 conversion records, each conversion record including: - A conversion photo representing a dental scene, and - A view or conversion view of a converted digital three-dimensional model that models the dental scene, the conversion view representing the scene in the same way as the conversion photo; 22) Train at least one conversion neural network through the conversion learning library so that the at least one conversion neural network learns how to convert any view of any digital three-dimensional model into a hyper-realistic view; 23) Submit the original view to the at least one conversion neural network so that the at least one conversion neural network converts the original view into a hyper-realistic view, wherein the hyper-realistic view is presented in the same way as a photo.

2. The conversion method according to claim 1, wherein In step 23), before submitting the original view to the conversion neural network, process the original view through a 3D engine.

3. A method for creating a hyper-realistic view of an original digital three-dimensional model, the method comprising the following steps: 21') Create a texture learning library that is unpaired or consists of more than 1000 texture records, each texture record including: - A non-realistic texture model representing a dental arch and a description of the non-realistic texture model, the description of the non-realistic texture model indicating that the non-realistic texture model is a non-realistic texture, or - A realistic texture model representing a dental arch and a description of the realistic texture model, the description of the realistic texture model indicating that the realistic texture model is a realistic texture; 22') Train at least one texture neural network through the texture learning library so that the at least one texture neural network learns to texture an initially non-textured model realistically; 23') Submit the original model to the at least one trained texture neural network so that the at least one trained texture neural network textures the original model to make it hyper-realistic; 24') Obtain a hyper-realistic view by observing the original model that has become hyper-realistic, wherein the hyper-realistic view is presented in the same way as a photo.

4. A method for enriching a historical learning library, the method comprising the following steps: 1) Generate an original digital three-dimensional model or a historical model of a dental arch; 2) Create a hyper-realistic view of the historical model from the original view of the historical model according to the method according to any one of claims 1 to 3; 3) Create a description or historical description of the hyper-realistic view; 4) Create a historical record consisting of the hyper-realistic view and the historical description, and add the historical record to the historical learning library, wherein the hyper-realistic view is presented in the same way as a photo.

5. The method according to claim 4, wherein, In step 1), generate a description of the historical model, and in step 3), create the historical description at least partially from the description of the historical model.

6. According to the method of claim 5, wherein, - Divide the historical model into a plurality of basic models, - In step 1), generate a specific description of the basic model to be represented in the hyper-realistic view in the description of the historical model, and - In step 3), include the specific description of the representation of the basic model in the hyper-realistic view in the historical description, at least a part of the specific description of the representation of the basic model in the hyper-realistic view is inherited from the specific description of the basic model, wherein the basic model is a model of teeth and / or tongue, and / or mouth, and / or lips, and / or jaw, and / or gums, and / or dental device.

7. The method according to any one of claims 4 to 6, including the following step 5) after step 4): 5) Modify the hyper-realistic view, and then return to step 3).

8. The method according to any one of claims 4 to 6, including the following step 6) after step 4): 6) Deform the historical model, and then return to step 1), Among them, wherein the deformation includes: - Move the tooth model to simulate the spacing between two teeth, - Deform the tooth model to simulate bruxism, - Delete the tooth model, - Deform the jaw model.

9. The method according to claim 8, wherein, Deform the historical model to represent a theoretical tooth situation.

10. The method according to claim 7, including the following step 6) after step 5): 6) Deform the historical model, and then return to step 1), Among them, wherein the deformation includes: - Move the tooth model to simulate the spacing between two teeth, - Deform the tooth model to simulate bruxism, - Delete the tooth model, - Deform the jaw model.

11. A method for analyzing an analysis photograph representing the dental arch of a patient, the method including the following steps: A) Create a historical learning library containing more than 1000 historical records by implementing the enrichment method according to any one of claims 4 to 10; B) Train at least one analysis neural network through the historical learning library, such that the at least one analysis neural network learns to evaluate the value of the attribute evaluated in the historical description for the photograph presented to it; C) Submit the analysis photograph to the trained neural network to obtain a description of the analysis photograph.

12. A method for simulating a tooth situation, the method including the following steps: A') At an update time, generate a digital three-dimensional model of the patient's dental arch, called the update model; B') Deform the update model to simulate the influence of the time between the update time and the simulation time to obtain a simulation model; C') Obtain a view of the simulation model or the original simulation view; D') Convert the original simulation view into a hyper-realistic simulation view according to the method according to claim 1.

13. The method according to claim 12, wherein, In step A'), divide the update model into multiple basic models, and in step B'), move and / or deform one or more basic models, wherein the movement and / or deformation includes: - Move the tooth model to simulate the spacing between two teeth, - Deform the tooth model to simulate bruxism, - Delete the tooth model, - Deform the jaw model.

14. A method for generating a hyper-realistic model from an original model of a dental arch, the method comprising the following consecutive steps: A”) Obtaining an original view of the original model; B”) Converting the original view into a hyper-realistic view according to the conversion method as claimed in claim 1; C”) For each pixel of the hyper-realistic view, identifying the corresponding voxel of the original model and assigning the attribute value of the pixel to the attribute of the voxel.

Images