3D face model construction method for face beauty and plastic surgery

Abstract

The invention discloses a 3D facial model construction method for facial cosmetic plastic surgery, relates to the technical field of plastic surgery model construction, and improves the scientificity and accuracy of plastic surgery decision. According to the method, all the face-lifting images are matched with the user face feature points, the face change prediction network is established according to the matching result, the user face model is established according to the face image data shot at all the angles, and the face change prediction network is mapped to the user face model; a plurality of face-lifting displacement vectors are generated on the face change prediction network according to the face cosmetic plastic surgery decision, the predicted change position of each user face feature point is obtained according to the face-lifting image associated with the user face feature point corresponding to each face-lifting displacement vector, and then the face cosmetic plastic surgery decision is adjusted synchronously. And adjusting the user face model according to the optimized facial cosmetic plastic surgery decision to obtain a 3D facial plastic surgery result model.

Description

Technical Field

[0001] The present invention relates to the technical field of cosmetic model construction, and specifically to a 3D facial model construction method for facial cosmetic plastic surgery. Background Art

[0002] In the field of cosmetic plastic surgery, traditional methods for predicting plastic surgery effects often have many limitations. In the past, doctors usually could only describe the approximate post-surgery effects to patients based on their own experience and limited two-dimensional pictures. This method made it difficult for patients to intuitively and accurately feel the specific changes in all angles after plastic surgery, resulting in a large deviation between the patient's expectation of the plastic surgery effect and the actual result, thus affecting the patient's satisfaction.

[0003] Moreover, traditional methods lack precise analysis of individual facial features and personalized effect simulation. The facial features of different patients vary greatly, and simply referring to general plastic surgery cases cannot meet the special needs of each patient. At the same time, during the plastic surgery decision-making process, the communication between doctors and patients is not efficient and precise enough, making it difficult to adjust the plastic surgery plan in real time according to the patient's expectations, resulting in the final plastic surgery plan may not be the most suitable for the patient.

[0004] With the development of technology, although some facial simulation technologies have emerged, most of these technologies can only perform simple two-dimensional simulations, unable to provide real and comprehensive three-dimensional plastic surgery effect displays, and also difficult to achieve dynamic and precise adjustments according to the specific facial features and plastic surgery needs of patients. Therefore, a 3D facial model construction method for facial cosmetic plastic surgery is provided. Summary of the Invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a 3D facial model construction method for facial cosmetic plastic surgery.

[0006] To achieve the above purpose, the present invention provides the following technical solutions: A 3D facial model construction method for facial cosmetic plastic surgery, comprising the following steps: Step S1: Take multi-angle photos of the user's face to generate facial image data, match the facial image data taken from each angle through the Internet, and then obtain several post-surgery images; Step S2: Extract several user facial feature points from the facial image data taken from each angle, match each post-surgery image with the user facial feature points, associate each post-surgery image with the user facial feature points according to the matching results, then establish a facial change prediction network, establish a user facial model based on the facial image data taken from each angle, and map the facial change prediction network onto the user facial model; Step S3: Obtain the facial cosmetic plastic surgery decision, generate a number of plastic surgery displacement vectors on the facial change prediction network according to the facial cosmetic plastic surgery decision, obtain the predicted change positions of each user facial feature point based on each plastic surgery displacement vector and the corresponding post-plastic surgery image of the user facial feature point, and then synchronously adjust the facial cosmetic plastic surgery decision; Step S4: Perform position changes on the user facial feature points on the user facial model according to the optimized facial cosmetic plastic surgery decision, synchronously adjust the entire user facial model, and then output the 3D facial plastic surgery result model corresponding to the facial cosmetic plastic surgery decision.

[0007] Further, the process of collecting the facial image data includes: Deploy 12 high-resolution cameras in a ring array in a standardized photography studio. Each high-resolution camera is evenly distributed around the user's head at 15° intervals, ensuring that the shooting range of the high-resolution cameras covers the front, side, top-down, and bottom-up views of the user's face. At the same time, there is partial overlap in the shooting ranges of adjacent high-resolution cameras arranged in sequential spatial positions; Synchronously trigger all high-resolution cameras to take pictures to obtain the facial image data of the user's face under natural light and a specific wavelength.

[0008] Further, the process of obtaining the post-plastic surgery image includes: Adopt an image retrieval algorithm based on deep hashing, use the facial image data taken at each angle as index materials, and quickly match the post-plastic surgery images with a similarity higher than 90% through the Internet; After the post-plastic surgery image matching is completed, establish a two-dimensional coordinate system, overlap and map each post-plastic surgery image and the facial image data onto the two-dimensional coordinate system, and then align the key points of the post-plastic surgery image with the current user facial key points by using affine transformation to eliminate the deviation caused by the shooting angle difference.

[0009] Further, the process of extracting a number of user facial feature points from the facial image data taken at each angle includes: Match and connect the facial image data taken at each angle to obtain the user facial image data, and establish a user facial model based on the user facial image data; Adopt a cascaded convolutional neural network to extract n user facial feature points from the facial image data taken at each angle, where n is a natural number greater than 0; Match the user facial feature points on the facial image data taken at each angle. If it is determined that two user facial feature points correspond to the same position, then merge the corresponding user facial feature points, otherwise do not perform any operation.

[0010] Further, the process of matching the post-plastic surgery image with the user facial feature points includes: Extract a number of front facial feature points and rear facial feature points from the pre- and post-plastic surgery images, and successively use each user's facial feature point as the central feature point to compare the front facial feature points on each pre- and post-plastic surgery image with the central feature point. The comparison process includes: With the central feature point as the center, select the nearest user facial feature point in the 360° direction centered on the central feature point at intervals of 10° as the associated user facial feature point. Then, match the front facial feature points on each pre- and post-plastic surgery image with the central feature point respectively, and obtain the spatial distance between each front facial feature point and the central feature point according to the matching result. Set a spatial distance threshold. If the spatial distance between the front facial feature point and the central feature point is less than or equal to the spatial distance threshold, then select the associated front facial feature point around the corresponding front facial feature point according to the process of selecting the associated user facial feature point with the central feature point. If the spatial distance between the front facial feature point and the central feature point is greater than the spatial distance threshold, do nothing.

[0011] Furthermore, the establishment process of the facial change prediction network includes: Taking the central feature point and the front facial feature point as the starting points, and each associated user facial feature point and the associated front facial feature point as the ending points, establish a number of user facial extension vectors and front facial extension vectors, match the user facial extension vectors and the front facial extension vectors with each other, and then obtain the product of the user facial extension vector and the front facial extension vector with the smallest included angle between the two. Accumulate the total number of products between the user facial extension vectors and the front facial extension vectors, denoted as the matching correlation degree, and set a correlation degree threshold. If there is a matching correlation degree between a front facial feature point and the central feature point that is greater than or equal to the correlation degree threshold, then associate the pre- and post-plastic surgery image where the front facial feature point is located with the corresponding central feature point, otherwise do nothing. Map each user facial feature point to the user facial model, and according to the result of matching the central feature point with the associated user facial feature point, connect each user facial feature point with its matched associated user facial feature point, thereby obtaining the facial change prediction network, and synchronously associate the pre- and post-plastic surgery images matched by each user facial feature point to the facial change prediction network.

[0012] Furthermore, the process of generating a number of plastic surgery displacement vectors on the facial change prediction network according to the facial cosmetic surgery decision includes: The facial cosmetic plastic surgery decision includes the predicted change positions of various parts on the user's facial model. A three-dimensional coordinate system is established, and the predicted change positions of various parts in the facial cosmetic plastic surgery decision are bound to the user's facial feature points on the user's facial model and mapped into the three-dimensional coordinate system. According to the facial cosmetic plastic surgery decision, the predicted spatial displacement positions of each user's facial feature point are obtained, and then a multi-dimensional vector D corresponding to the facial cosmetic plastic surgery decision is generated; The multi-dimensional vector D = {d1, d2, ······, d n}, where d n represents the plastic surgery displacement vector of the nth user's facial feature point.

[0013] Furthermore, the process of adjusting the facial cosmetic plastic surgery decision includes: Establish an optimization objective function, taking into account aesthetic scores and physiological feasibility:

[0014] where and respectively represent the aesthetic loss of the golden ratio and the muscle and bone constraint loss, and represent constant correction parameters, represents the coordination value; According to the post-plastic surgery images associated with each user's facial feature point, several associated displacement vectors are set for the plastic surgery displacement vector of each user's facial feature point in the multi-dimensional vector D. The angle between each associated displacement vector and the corresponding plastic surgery displacement vector is within (-90°, 90°), and the included angles between each associated displacement vector and the corresponding plastic surgery displacement vector are all different. At the same time, the starting point of the associated displacement vector is the user's facial feature point, but the ending positions are all located on the post-plastic surgery image; That is, each associated displacement vector comes from the front and back position changes of the position associated with the user's facial feature point in the post-plastic surgery image; Input each associated displacement vector and the plastic surgery displacement vector into the optimization objective function. Then, the optimization objective function outputs the associated displacement vector or the plastic surgery displacement vector with the smallest coordination value, and the output displacement vector is recorded as the predicted change position of the corresponding user's facial feature point; When the predicted change positions of all user's facial feature points are updated, the correction of the facial cosmetic plastic surgery decision is completed.

[0015] Furthermore, the generation process of the 3D facial plastic surgery result model includes: Map the predicted change positions of each user's facial feature point in the facial cosmetic plastic surgery decision to the facial change prediction network on the user's facial model, and then use the Laplace smoothing algorithm to eliminate the deformation of the user's facial model caused by the change of the user's facial feature point position: After the position changes and corrections of all the user's facial feature points are completed according to the facial cosmetic plastic surgery decision, the 3D facial plastic surgery result model corresponding to the facial cosmetic plastic surgery decision is output.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Extract the user's facial feature points from the facial image data, match and associate the post-plastic surgery images therewith, establish a facial change prediction network and map it onto the user's facial model. This method fully considers the unique facial features of each user, can provide highly personalized plastic surgery effect prediction for users, and meet the special needs of different users.

[0017] 2. The present invention generates a plastic surgery displacement vector according to the facial cosmetic plastic surgery decision, adjusts it, and synchronously adjusts the facial cosmetic plastic surgery decision. This enables dynamic adjustment in real time according to the user's facial features and expectations during the plastic surgery decision-making process, ensuring that the final plastic surgery plan is the most suitable for the user, and improving the scientificity and accuracy of the plastic surgery decision. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention.

[0019] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other implementation manners obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope protected by the present invention.

[0021] As Figure 1 shown, a 3D facial model construction method for facial cosmetic plastic surgery includes the following steps: Step S1: Take multiple-angle photos of the user's face to generate facial image data, match the facial image data taken from each angle through the Internet, and thus obtain several post-plastic surgery images; Step S2: Extract several user facial feature points from the facial image data captured from various angles, match each post-plastic surgery image with the user facial feature points, associate each post-plastic surgery image with the user facial feature points according to the matching results, and then establish a facial change prediction network. Establish a user facial model based on the facial image data captured from various angles, and map the facial change prediction network onto the user facial model; Step S3: Obtain a facial beauty plastic surgery decision, generate several plastic surgery displacement vectors on the facial change prediction network according to the facial beauty plastic surgery decision, obtain the predicted change positions of each user facial feature point corresponding to the post-plastic surgery image associated with the user facial feature point according to each plastic surgery displacement vector, and then adjust the facial beauty plastic surgery decision synchronously; Step S4: Change the positions of the user facial feature points on the user facial model according to the optimized facial beauty plastic surgery decision, and adjust the entire user facial model synchronously, and then output a 3D facial plastic surgery result model corresponding to the facial beauty plastic surgery decision.

[0022] Further, the above Step S1 is implemented through the following steps: Deploy 12 high-resolution cameras (resolution not less than 24 million pixels) in a ring array in a standardized photography studio. Each high-resolution camera is evenly distributed around the user's head at an interval of 15°, ensuring that the shooting range of the high-resolution cameras covers the front, side, top-down and bottom-up views of the user's face, and at the same time, there is partial overlap in the shooting ranges of the high-resolution cameras distributed in adjacent sequential spatial positions; Trigger all high-resolution cameras to take pictures synchronously, and obtain facial image data of the user's face under natural light and a specific wavelength (such as UV light). The facial image data includes facial structure features such as the user's skin texture, pore distribution and subcutaneous blood vessels; Adopt an image retrieval algorithm based on deep hashing, use the facial image data captured from various angles as index materials, and quickly match post-plastic surgery images with a similarity higher than 90% through the Internet. The hashing algorithm formula is: ; where represents the result of image feature analysis, W is the hashing weight matrix, I represents the facial image data, and represent pooling operation and convolution operation respectively, and T represents the matrix dimension; When the matching of the post-plastic surgery images is completed, establish a two-dimensional coordinate system, overlap and map each post-plastic surgery image and the facial image data onto the two-dimensional coordinate system, and then align the key points (such as the corners of the eyes and the tip of the nose) of the post-plastic surgery image with the current facial key points of the user by using affine transformation to eliminate the deviation caused by the shooting angle difference.

[0023] Further, the step S2 is implemented through the following steps: Match and connect the facial image data captured from various angles. Since all the facial image data correspond to the same user, there must be identical parts in the facial image data. Then, splice the facial image data from various angles to obtain the user's facial image data, and establish a user facial model based on the user's facial image data; Extract n user facial feature points from the facial image data captured from various angles. The user facial feature points include anatomical landmarks such as the supraorbital ridge, eyelids, alae nasi, and corners of the mouth. The extraction formula is: ; Where represents the i-th user facial feature point, represents the shooting angle, represents the convolution operation, I represents the facial image data, i is a natural number less than or equal to n, and n is a natural number greater than 0; Match the user facial feature points on the facial image data captured from various angles. If it is determined that two user facial feature points correspond to the same position, then merge the corresponding user facial feature points; otherwise, do nothing. During the process of extracting user facial feature points from the facial image data, extract several front facial feature points and rear facial feature points from the pre- and post-plastic surgery images. Then, taking each user facial feature point as the central feature point in turn, compare the front facial feature points on each pre- and post-plastic surgery image with the central feature point. The comparison process includes: Taking the central feature point as the center, select the nearest user facial feature point in the 360° direction centered on the central feature point at intervals of 10° as the associated user facial feature point. Then, match the front facial feature points on each pre- and post-plastic surgery image with the central feature point respectively, and obtain the spatial distance between each front facial feature point and the central feature point according to the matching result; Set a spatial distance threshold. If the spatial distance between the front facial feature point and the central feature point is less than or equal to the spatial distance threshold, then select the associated front facial feature point around the corresponding front facial feature point according to the process of selecting the associated user facial feature point based on the central feature point; If the spatial distance between the front facial feature point and the central feature point is greater than the spatial distance threshold, then do nothing; Taking the central feature point and the front facial feature point as the starting points, and each associated user facial feature point and the associated front facial feature point as the ending points, establish several user facial extension vectors and front facial extension vectors, match the user facial extension vectors and the front facial extension vectors with each other, and then obtain the product of the user facial extension vector and the front facial extension vector with the smallest included angle between them; Accumulate the total product between the facial extension vectors of each user and the previous facial extension vectors, denoted as the matching correlation degree, and set a correlation degree threshold. If there is a matching correlation degree between a previous facial feature point and the central feature point greater than or equal to the correlation degree threshold, then associate the pre- and post-cosmetic images where the previous facial feature point is located with the corresponding central feature point; otherwise, do nothing. Map each user's facial feature points to the user's facial model, and based on the result of matching and associating the user's facial feature points with the central feature point, connect each user's facial feature points with their matching associated user's facial feature points to obtain a facial change prediction network, and synchronously associate the pre- and post-cosmetic images matched by each user's facial feature points to the facial change prediction network.

[0024] Further, step S3 is implemented through the following steps: The doctor uploads a facial cosmetic surgery decision, which includes the predicted change positions of various parts on the user's facial model, such as nose bridge elevation, mandibular angle reduction, etc.; Establish a three-dimensional coordinate system, bind the predicted change positions of various parts in the facial cosmetic surgery decision to the user's facial feature points on the user's facial model and map them into the three-dimensional coordinate system, and obtain the predicted spatial displacement positions of each user's facial feature points according to the facial cosmetic surgery decision, thereby generating a multi-dimensional vector D corresponding to the facial cosmetic surgery decision; The multi-dimensional vector D = {d1, d2, ······, d n}, where d n represents the cosmetic displacement vector of the nth user's facial feature point, n ≥ 1; Establish an optimization objective function, taking into account aesthetic score and physiological feasibility:

[0025] where and respectively represent the aesthetic loss of the golden ratio and the muscle and bone constraint loss, and represent constant correction parameters, represents a coordination value; According to the post-cosmetic images associated with each user's facial feature points, set m associated displacement vectors for the cosmetic displacement vector of each user's facial feature point in the multi-dimensional vector D. The angle between each associated displacement vector and the corresponding cosmetic displacement vector is in (-90°, 90°), and the angles between each associated displacement vector and the corresponding cosmetic displacement vector are all different. At the same time, the starting point of the associated displacement vector is the user's facial feature point, but the end point positions are all on the post-cosmetic image, and m is a natural number greater than 0; That is, each associated displacement vector comes from the position change before and after the position associated with the user's facial feature points in the post-cosmetic image; Input each associated displacement vector and the cosmetic displacement vector into the optimization objective function. Then, the optimization objective function outputs the associated displacement vector or the cosmetic displacement vector with the smallest coordination value, and record the output displacement vector as the predicted change position of the corresponding user's facial feature points; After the predicted change positions of all user's facial feature points are updated, the correction of the facial cosmetic surgery decision is completed.

[0026] Further, the step S4 is implemented through the following steps: Map the predicted position changes of each user's facial feature points in the facial cosmetic surgery decision to the facial change prediction network on the user's facial model, and then use the Laplace smoothing algorithm to eliminate the deformation of the user's facial model caused by the change of the position of the user's facial feature points: ; where represents the coordinate position of the changed point of the associated user's facial feature points of the i-th user's facial feature points, represents the coordinate position of the t-th and i-th user's facial feature points according to the predicted change position in the facial cosmetic surgery decision, represents the corrected coordinate position of the i-th user's facial feature points, is the smoothing intensity parameter, ∈(0, 1); After the position changes and corrections of all user's facial feature points are completed according to the facial cosmetic surgery decision, output the 3D facial cosmetic result model corresponding to the facial cosmetic surgery decision.

[0027] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A 3D facial model construction method for facial cosmetic plastic surgery, characterized in that, Including the following steps: Step S1: Take multi-angle photos of the user's face to generate facial image data, match the facial image data taken at each angle through the Internet, and then obtain several post-plastic surgery images; Step S2: Extract several user facial feature points from the facial image data taken at each angle, match each post-plastic surgery image with the user facial feature points, associate each post-plastic surgery image with the user facial feature points according to the matching results, and then establish a facial change prediction network. Establish a user facial model based on the facial image data taken at each angle, and map the facial change prediction network onto the user facial model; Step S3: Obtain a facial beauty plastic surgery decision, generate several plastic surgery displacement vectors on the facial change prediction network according to the facial beauty plastic surgery decision, and obtain the predicted change positions of each user facial feature point corresponding to the post-plastic surgery image associated with the user facial feature point according to each plastic surgery displacement vector, and then synchronously adjust the facial beauty plastic surgery decision; Step S4: Perform position changes on the user facial feature points on the user facial model according to the optimized facial beauty plastic surgery decision, and synchronously adjust the entire user facial model, and then output a 3D facial plastic surgery result model corresponding to the facial beauty plastic surgery decision.

2. The 3D facial model construction method for facial cosmetic plastic surgery according to claim 1, characterized in that, The process of collecting the facial image data includes: Deploy high-resolution cameras arranged in a circular array in a standardized photography studio. Each high-resolution camera is evenly distributed around the user's head at an interval of 15°, ensuring that the shooting range of the high-resolution camera covers the front, side, top-down, and bottom-up views of the user's face. At the same time, there is partial overlap in the shooting ranges of adjacent high-resolution cameras arranged in sequential spatial positions, and all high-resolution cameras are triggered synchronously to shoot the facial image data of the user's face.

3. A 3D facial model construction method for facial cosmetic plastic surgery according to claim 2, characterized in that, The process of obtaining the post-plastic surgery images includes: Using the facial image data taken at each angle as index materials, quickly match post-plastic surgery images with a similarity higher than 90% through the Internet. After the post-plastic surgery images are matched, establish a two-dimensional coordinate system, overlap and map each post-plastic surgery image and the facial image data onto the two-dimensional coordinate system, and then align the key points of the post-plastic surgery image with the current facial key points of the user by using affine transformation to eliminate the deviation caused by the shooting angle difference.

4. A 3D facial model construction method for facial cosmetic plastic surgery according to claim 3, characterized in that, The process of extracting several user facial feature points from the facial image data taken at each angle includes: Establish a user facial model based on the facial image data taken at each angle, and use a cascaded convolutional neural network to extract n user facial feature points from the facial image data taken at each angle, where n is a natural number greater than 0; Match the user facial feature points on the facial image data taken at each angle with each other. If it is determined that two user facial feature points correspond to the same position, the corresponding user facial feature points are merged, otherwise no operation is performed.

5. A 3D facial model construction method for facial cosmetic plastic surgery according to claim 4, characterized in that, The process of matching the post-plastic surgery images with the user facial feature points includes: Extract several pre-facial feature points and post-facial feature points from the pre- and post-plastic surgery images, and sequentially use each user facial feature point as the central feature point; Taking the central feature point as the center, in the 360° direction centered on it, select the nearest user facial feature point at intervals of 10° as the associated user facial feature point. Then, match the front facial feature points on each pre- and post-plastic surgery image with the central feature point respectively, and obtain the spatial distance between each front facial feature point and the central feature point according to the matching result; Set a spatial distance threshold. According to the size relationship between the spatial distance between the front facial feature point and the central feature point and the spatial distance threshold, select the associated front facial feature point around the corresponding front facial feature point.

6. A 3D facial model construction method for facial cosmetic plastic surgery according to claim 5, characterized in that, The establishment process of the facial change prediction network includes: Taking the central feature point and the front facial feature point as the starting points, and each associated user facial feature point and the associated front facial feature point as the end points, establish several user facial extension vectors and front facial extension vectors. Match each user facial extension vector and the front facial extension vector with each other, and then obtain the product of the user facial extension vector and the front facial extension vector with the smallest included angle between them; Accumulate the total number of products between each user facial extension vector and the front facial extension vector, which is denoted as the matching correlation degree. Set a correlation degree threshold. If there is a matching correlation degree between a front facial feature point and the central feature point that is greater than or equal to the correlation degree threshold, then associate the pre- and post-plastic surgery image where the front facial feature point is located with the corresponding central feature point, otherwise do nothing; Map each user facial feature point to the user facial model, and according to the result of matching the central feature point with the associated user facial feature point, connect each user facial feature point with its matching associated user facial feature point, thereby obtaining the facial change prediction network, and synchronously associate the pre- and post-plastic surgery images matched by each user facial feature point to the facial change prediction network.

7. A 3D facial model construction method for facial cosmetic plastic surgery according to claim 6, characterized in that, The process of generating several plastic surgery displacement vectors on the facial change prediction network according to the facial cosmetic plastic surgery decision includes: The facial cosmetic plastic surgery decision includes the predicted change positions of each part on the user facial model. Establish a three-dimensional coordinate system, bind the predicted change positions of each part in the facial cosmetic plastic surgery decision to the user facial feature points on the user facial model and map them into the three-dimensional coordinate system. Obtain the predicted spatial displacement positions of each user facial feature point according to the facial cosmetic plastic surgery decision, and then generate the multi-dimensional vector D corresponding to the facial cosmetic plastic surgery decision.

8. A 3D facial model construction method for facial cosmetic plastic surgery according to claim 7, characterized in that, The process of adjusting the facial cosmetic plastic surgery decision includes: Establish an optimization objective function. According to the post-plastic surgery images associated with each user facial feature point, set several associated displacement vectors for the plastic surgery displacement vector of each user facial feature point in the multi-dimensional vector D; Input each associated displacement vector and the plastic surgery displacement vector into the optimization objective function. Then, the optimization objective function outputs the associated displacement vector or the plastic surgery displacement vector with the smallest coordination value, and record the output displacement vector as the predicted change position of the corresponding user facial feature point; When the predicted change positions of all user facial feature points are updated, the correction of the facial cosmetic plastic surgery decision is completed.

9. A 3D facial model construction method for facial cosmetic plastic surgery according to claim 8, characterized in that, The generation process of the 3D facial plastic surgery result model includes: A facial change prediction network that maps the predicted change positions of each user's facial feature points in facial cosmetic surgery decisions onto the user's facial model, and then uses the Laplacian smoothing algorithm to eliminate the deformation of the user's facial model caused by the change in the positions of the user's facial feature points: When all the position changes and corrections of the user's facial feature points are completed according to the facial cosmetic surgery decision, a 3D facial plastic surgery result model corresponding to the facial cosmetic surgery decision is output.

Images