A multi-layer garment transfer method preserving design semantics
By acquiring clothing skeleton information and using energy optimization technology, the problems of abrupt changes in skeleton correspondence and clipping in 3D clothing migration were solved, achieving high-quality migration of multi-layered clothing between different human bodies and improving the accuracy and realism of the migration results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2023-03-15
- Publication Date
- 2026-05-29
AI Technical Summary
Existing 3D clothing transfer methods struggle to handle the correspondence between clothing and skeleton when transferring between different human bodies, especially in complex poses. Furthermore, the clipping of multiple layers is difficult to resolve effectively, leading to distorted transfer results.
By acquiring the skeletal information corresponding to the clothing, the initial skinning result is generated using the biharmonic interpolation method. Combined with energy optimization technology, the clothing normal and human position are optimized to eliminate the influence of molding and posture differences. The ADMM method is used to optimize the clothing mesh to achieve accurate transfer of multi-layer clothing.
It achieves accuracy and realism in the transfer of multi-layer clothing between different human bodies, improving the transfer effect, especially when there are large differences in body size, the transfer results and the wearing effect of the original clothing on the new body are significantly improved.
Smart Images
Figure CN116524161B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of 3D clothing model transfer in computer graphics, and particularly relates to a multi-layer clothing transfer method that preserves design semantics. Background Technology
[0002] In recent years, digital clothing has seen widespread demand in fields such as virtual reality, digital entertainment, and the cosmos. By combining realistic folds with human models in specific poses, it achieves a strong sense of realism. However, creating high-quality digital clothing models is a labor-intensive task. In practice, there is a need to transfer clothing between different human bodies with similar poses, but currently there is no fully automated method for transferring multiple layers of clothing between different human bodies in similar poses.
[0003] While current methods for 3D clothing transfer using skeletal control can transfer between different human bodies in relatively simple poses, distortion inevitably occurs at the skeletal junctions. Furthermore, they struggle to handle the correspondence between clothing and skeletons in complex poses and are difficult to effectively resolve clipping issues between multiple layers of a single garment. Summary of the Invention
[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a multi-layer clothing transfer method that preserves design semantics.
[0005] The objective of this invention is achieved through the following technical solution:
[0006] A multi-layered clothing transfer method that preserves design semantics includes the following steps:
[0007] (1) Obtain the skeletal information corresponding to the clothing:
[0008] The skinning weights corresponding to the character model are obtained using the skeletal model corresponding to the character model, and a threshold is set for the weights. The clothing area corresponding to each bone is found by identifying the body area with a weight greater than the threshold for each bone. For a body point p on a bone with a weight greater than the threshold... s0 Find the point p on the skeletal line segment that is far from the body. s0 The nearest point p b0 , with p b0 Starting point Find clothing points along the direction, and record the point with the largest centroid coordinate weight of the triangle containing the found clothing point as p. g All p found g Recorded in a collection middle;
[0009] (2) Initial migration of clothing with corresponding skeletal structures:
[0010] For a set p in g Find the distance p on that bone g The nearest point p b and find the line segment The intersection point p with the body model s ;The point p on the human body model and skeleton after the transfer is obtained by utilizing the correspondence between the human body model before and after the transfer. s ′ and p′ b The new clothing point p′ is calculated using the following formula. g :
[0011]
[0012] (3) Sampling multiple layers and closely spaced areas:
[0013] Uniformly sample the vertices of the mesh on the clothing, and obtain a point after each sampling. i satisfies 0 ≤ i ≤ p·v s Where i represents the sampling point number, p represents the sampling rate of uniform sampling, and v s This represents the total number of vertices in the clothing model;
[0014] from Starting from point d, in its normal direction and the opposite direction of the normal direction, at a distance of threshold d max All points obtained by the intersection of the inner part and its clothing are denoted as Construct its point-to-point sequence And record all the sequences obtained from sampling;
[0015] (4) Initial migration of clothing points that do not have a corresponding skeletal point relationship:
[0016] Using clothing points with corresponding skeletal relationships as constraints, and leveraging the continuity of the original clothing mesh, biharmonic interpolation is used to obtain preliminary migration results for clothing points without skeletal relationships. Simultaneously, the constraint step (3) ensures that the displacements of point pairs in the point pair sequence are equal, thus obtaining the preliminary skinning result p of the migration of all points including unconstrained points. skinning .
[0017] (5) Using clothing points with skeletal relationships as constraints, interpolate to obtain the local coordinate system corresponding to clothing points without skeletal relationships:
[0018] Export the local coordinate system and coordinate axes of each bone in the human model before and after migration from the Autorig bone controller plugin. Simultaneously calculate the Euler angles from the original coordinate system to the local coordinate axes of each bone. The Euler angles corresponding to the human bones before migration are denoted as follows: The Euler angles of the human skeleton of the migration target are denoted as By utilizing the continuity of the clothing mesh and using biharmonic interpolation, the Euler angles of the skeleton coordinate system corresponding to the clothing point are used as the boundary to obtain the Euler angles of the local coordinate system corresponding to all points including unconstrained points; the transformation matrix of rotating the original coordinate system according to the Euler angles corresponding to the clothing point before migration is denoted as τ; the transformation matrix of rotating the Euler angles corresponding to the clothing point of the migration target is denoted as τ′.
[0019] (6) Transform the point pairs in step (3) into linear constraints in energy optimization:
[0020] For a set of point pairs in the point pair sequence obtained in step (3) <p a ,p b >, construct the following formula (2):
[0021]
[0022] in This indicates that point a is at p. skinning The same applies to the y and z coordinates in the results for the x-coordinates. Represents line segment The length, ∈ slack For the relaxation constant at this point, 0.3 is used in practice; the x, y, z coordinates of all point pairs obtained in step (3) are constrained as in formula (3) and added to the linear constraint matrix R.
[0023] (7) Construct the following energy function for each point on the garment:
[0024] After obtaining preliminary migration skinning results p skinning Then, a 3×3 matrix representation is constructed for the triangular facets before and after optimization. This matrix is denoted as P on the original garment. P is defined as follows: t = (p4-p1, p4-p2, p4-p3);
[0025] Where p1, p2, and p3 are column vectors representing the coordinates of the three vertices of the triangle. in This represents the normal vector of the triangle;
[0026] Similarly, a 3×3 matrix is constructed on the target garment to be optimized, and recorded as follows: The gradient of each triangle face transformation in the clothing migration is then expressed as: Then, the energy of each triangular facet is accumulated to obtain the conformal energy E. shape Then update the energy components in sequence.
[0027] (8) Eliminate clipping between clothing and the human body:
[0028] The clipping of clothing and human body is divided into clipping between clothing points and human body models and clipping between clothing surfaces and human body models, and then these two types of states are judged respectively.
[0029] Furthermore, in step (7), the energy accumulation is used to obtain the conformal energy E. shape The mathematical expression is:
[0030]
[0031] Among them ||·|| F Let T be the norm of the matrix, i.e., Frobenius, T t For an ideal transformation that only involves transformations within the triangular face, i.e., in T t The transformation preserves the normal of the triangular facet; then the energy is updated in the following order:
[0032] Furthermore, in step (7), the energy components are updated in the following order:
[0033] (7.1) Update the T matrix corresponding to each facet in the current state; first, project each migrated triangular facet onto the plane where the migrated facet is located to obtain p'; the process is shown in equation (4) below:
[0034]
[0035] This yields the 3×3 matrix representation of the projected triangle, which is P′. t = (p′4-p′1, p′4-p′2, p′4-p′3), when satisfying Update T at that time t =P′ t (P t ) -1 Otherwise, keep T constant, so that the energy decreases when updating T;
[0036] (7.2) Construct E shape Construct shape-preserving energy according to formula (3);
[0037] (7.3) Constructing the anti-torsional energy E rl For clothing points corresponding to skeletal points in step (2), construct the energy as shown in equation (5):
[0038]
[0039] Where α is a constant, set as Represents the total points of the clothing. <p ref >Represents the total number of clothing points with skeletal markings in step (2); The vector representing the connection between the current optimization result and the skinning result obtained in step (4); d b The direction of the bone corresponding to the current point is represented by d. p Indicates and (p) b ,p skinning If the unit vectors are collinear, then d t =d b ×d p Through E rl Can be kept In (p) b ,p skinning While restricting the directional degree of freedom, it also restricts movement in a plane orthogonal to it, thereby preventing torsion.
[0040] (7.4) For the p area marked as needing to be closely attached fit The energy E for constructing the tightly attached portion is as described in equation (6). fit :
[0041]
[0042] (7.5) Substitute the current constraint term R into the equation and use the ADMM method to optimize E = E while satisfying the R constraint. shape +E rl +E fit The quadratic programming problem is solved to obtain the final optimized clothing mesh result, denoted as p. energy .
[0043] Furthermore, step (8) specifically includes the following sub-steps:
[0044] (8.1) The first state is where there is clipping between clothing points and the human body model: For each point on the clothing, find the point on the human body model that is closest to this point, and at the same time determine whether this point is inside the human body model based on the normal of the triangle containing this point; all points p inside the human body model collision And the nearest clothing point g closest and the corresponding human face normal. Add the constraint shown in equation (7) in R:
[0045]
[0046] (8.2) The second state is when there is no clipping between the clothing point and the human body model, and an intersection between the human body model and the clothing model is detected: for each set of intersecting triangles, point p collision And the nearest clothing point g closest And the normal direction of this human body. Add the constraints shown in equation (7) in R;
[0047] Repeat steps (7) and (8) until there are no new constraints in the last step (8) and the energy of step (7) converges, thus obtaining the result of eliminating clipping between the character and clothing.
[0048] The beneficial effects of this invention are as follows: This invention utilizes the correspondence between the human body model and the human skeleton before and after migration. It first obtains the initial skinning result using biharmonic interpolation, and then optimizes the clothing normal using energy optimization. This solves the problems of the impact of abrupt changes in the skeleton correspondence during skinning and the discrepancies between clothing fit and human posture during energy optimization. This clothing migration method does not use any physically relevant parameters other than clothing geometry information. A comparison between this clothing migration method and the results of manual sculpting proves the correctness of the reconstruction results. In clothing migration, the migration results can reflect the effect of the original clothing on the new human body, and the effect is significantly improved compared to the original method when there is a large difference in body size before and after migration. Attached Figure Description
[0049] Figure 1 Here is a flowchart of the skinning process;
[0050] Figure 2 Flowchart for energy optimization section;
[0051] Figure 3 The images show clothing and human body models at different stages of a clothing migration process; where (a) is the human body model and clothing from the migration source, (b) is the clothing and target human body model from the intermediate migration result, and (c) is the clothing and human body model from the migration result.
[0052] Figure 4 A close-up of a cuff for two migrations with the same migration source and migration target, where (a) no skeletal local coordinate system transformation was used, and (b) a skeletal local coordinate system transformation was used;
[0053] Figure 5 This is an example diagram of a migration involving multiple garments, where (a) represents the source garment and human body model, and (b) represents the result garment and human body model. Detailed Implementation
[0054] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0055] The general process of this invention is as follows: first, a preliminary skinning migration result is generated based on the skeletal relationship corresponding to the clothing; then, the clothing normal and its relative position to the human body are optimized simultaneously. The specific process of this invention will now be described in detail.
[0056] Step 1: Obtain the skeletal information corresponding to the clothing
[0057] Skeletal points are marked on a standard human body model with the same topology as the original human body model, and each skeletal point is connected to several other skeletal points to form a skeleton. A weight matrix W is then calculated using median value coordinates. n,m {w i,j}, where n and m are the number of skeletal points and the number of vertices in the human body mesh, respectively, and w i,j Let v represent the value in the i-th row and k-th column of the matrix, satisfying equation (1), where v k ,b i Let W represent the coordinates of the j-th vertex and the i-th bone point of the human body mesh, respectively. For the original human body model to be transferred, W... n,m By applying the same method to all vertices in S, the same number of skeletal points can be obtained.
[0058] ∑ k w i,k v k =b i ′ (1)
[0059] After obtaining the skeletal information of the human body, the skinning weights provided by the SMPL model are used, and a threshold is set for the weights. The clothing area corresponding to each bone is found by identifying the body area whose weight is greater than the threshold for each bone. For a body point p on a bone with a weight greater than the threshold... s0 Find the point p that is closest to the given point on the skeletal line segment. b0 Then with p b0 Starting point Find clothing points along the direction, and record the point with the largest centroid coordinate weight of the triangle containing the found clothing point as p. g .
[0060] Step 2: Perform preliminary migration of clothing with corresponding skeletal structures.
[0061] For a p that corresponds to a skeleton g Find the distance p on that bone g The nearest point p b and find the line segment The intersection point p with the body model s By utilizing the correspondence between the human body models before and after migration, the point p on the human body model and skeleton after migration is obtained. s ′ and p′ b The new clothing point p′ is calculated using the following formula. g :
[0062]
[0063] Step 3: Sample multiple layers and closely spaced areas.
[0064] To address the issue of maintaining relative positions between multiple layers and tightly fitting parts within a single garment, point sampling is performed on the garment. For each sampled point... Starting from that point, the distance between its normal and the opposite direction will be at a distance threshold d. max The point where the inner part intersects with the garment is denoted as . Construct a point pair sequence for these points. All samples are recorded as sequences.
[0065] Step 4: Initial migration of clothing points that do not have a corresponding skeletal point relationship.
[0066] Using clothing points with corresponding skeletal relationships as constraints, and leveraging the continuity of the original clothing mesh, biharmonic interpolation is used to obtain preliminary migration results for clothing points without skeletal relationships. Simultaneously, the constraint in the third step ensures that the displacements of point pairs in the point pair sequence are equal, resulting in... Figure 4 The preliminary migration skinning results p for all points including unconstrained points are shown. skinning .
[0067] Step 5: Using clothing points with skeletal relationships as constraints, interpolate to obtain the local coordinate system corresponding to clothing points without skeletal relationships.
[0068] Export the local coordinate system axes of each bone in the human model before and after migration from the Autorig plugin. Simultaneously calculate the Euler angles from the original coordinate system to the local coordinate axes of each bone. The Euler angles corresponding to the human bones before migration are denoted as follows: The Euler angles of the human skeleton of the migration target are denoted as By leveraging the continuity of the clothing mesh and using biharmonic interpolation, with the Euler angles of the skeletal coordinate system corresponding to the clothing points as the boundary, we obtain the Euler angles of the local coordinate system corresponding to all points, including unconstrained points. The transformation matrix for rotating the original coordinate system according to the Euler angles corresponding to the clothing points before migration is denoted as τ; the transformation matrix for rotating the clothing points of the migration target according to the Euler angles is denoted as τ′.
[0069] Step 6: Transform the point pairs from Step 3 into linear constraints for energy optimization.
[0070] For a set of point pairs in the point pair sequence obtained in step three <p a ,p b >, construct the following formula (3):
[0071]
[0072] in This indicates that point a is at p. skinning The same applies to the y and z coordinates in the results for the x-coordinates. Represents line segment The length, ∈ slack In practice, the relaxation constant for this location is taken as 0.3.
[0073] For all the point pairs obtained in the third step, construct constraints as shown in (3) and add them to the linear constraint matrix R.
[0074] Step 7: Construct the energy function for each point on the garment according to the following method. This will yield preliminary migration skinning results. skinning Then, a 3×3 matrix representation is constructed for the triangular facets before and after optimization. This matrix is denoted as P on the original garment. P is defined as follows: t = (p4-p1, p4-p2, p4-p3), where p1, p2, and p3 are column vectors representing the coordinates of the three vertices of the triangle. in This represents the normal vector of the triangle. Similarly, a 3×3 matrix is constructed on the target garment to be optimized, denoted as... This allows us to express the gradient of each triangle face transformation during clothing migration as follows: The conformal energy E is obtained by summing the energies of each triangular facet as follows. shape :
[0075]
[0076] Among them ||·|| F Let T be the Frobenius norm of the matrix. t For an ideal transformation that only involves transformations within the triangular face, that is, in T t The transformation preserves the normal of the triangular facet. Then, the energy components are updated in the following order.
[0077] (a) Update the T matrix corresponding to each facet in the current state. First, project each migrated triangular facet onto the plane where the migrated facet is located to obtain p'. The process is shown in equation (4) below:
[0078]
[0079] This yields the 3×3 matrix representation of the projected triangle, which is P. ′t = (p′4-p′1, p′4-p′2, p′4-p′3), when satisfying Update T at that time t =P′ t (P t ) -1 Otherwise, keep T unchanged, which ensures that energy decreases when updating T.
[0080] (b) Constructing E shape Construct shape-preserving energy according to equation (4).
[0081] (c) Constructing a torsional energy E rl For clothing points corresponding to skeletal points in the second step, the energy is constructed as shown in equation (6):
[0082]
[0083] Where α is a constant, set as Represents the total points of the clothing.
[0084] <p ref > represents the total number of clothing points with skeletal markers in the second step. d represents the vector connecting the current optimization result and the skinning result obtained in the fourth step. b The direction of the bone corresponding to the current point is represented by d. p Indicates and (p) b ,p skinning If the unit vectors are collinear, then d t =d b ×d p Through E rl Can be kept In (p) b ,p skinning While restricting the directional degree of freedom, it also restricts movement in a plane orthogonal to it, thereby preventing torsion.
[0085] (d) For p that is marked as needing to be closely attached fit The energy E of the tightly packed portion is constructed as described in the following formula. fit :
[0086]
[0087] (f) Substitute the current constraint term R into the equation and use the ADMM method to optimize E = E while satisfying the R constraint. shape +E rl +E fit The quadratic programming problem is solved as follows: Figure 5 p shown energy .
[0088] Step 8: Eliminate clipping between clothing and the human body. Clipping between clothing and the human body can be divided into two types: the first is clipping between clothing points and the human body model, and the second is clipping between surfaces and the human body model. This step will determine these two types separately.
[0089] This step has two states:
[0090] The first scenario involves clipping between clothing points and the human body model. In this scenario, the first step is to resolve the clipping issue. For each point on the clothing, find the closest point on the human body model and determine if that point is inside the human body model based on the normal direction of the triangle containing that point. All points p that are inside the human body model... collision And the nearest clothing point g closest and the corresponding human face normal. Add the constraint in R as shown in equation (8):
[0091]
[0092] The second state is when there is no clipping between the clothing point and the human body model, but an intersection between the human body model and the clothing model is detected. For each pair of intersecting triangles, point p... collision And the nearest clothing point g closest And the normal direction of this human body. Add constraints to R as shown in equation (8).
[0093] Repeat steps seven and eight until there are no new constraints in the final step eight and the energy of step seven converges, thus obtaining... Figure 5 The results are shown.
[0094] Step 9: Eliminate clipping between multiple garments. For clipping between garments, a method from the outside in is used. First, migrate the outermost garments, then sample the inner garments when migrating them. Start sampling from point p on the inner garment. inside Starting along the normal direction of the inner garment at that point Find the first intersection point p with the outer garment outside Record such a set of point pairs as <p inside ,p outside For each pair of points, add the following constraints to R to obtain the final complete set of transfer results:
[0095]
[0096] The above-described embodiments illustrate specific implementations of the present invention. Their detailed descriptions are intended to aid in understanding the method and core ideas of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.
Claims
1. A multi-layered clothing transfer method that preserves design semantics, characterized in that, Includes the following steps: (1) Obtain the skeletal information corresponding to the clothing: The skinning weights corresponding to the character model are obtained using the skeletal model corresponding to the character model, and a threshold is set for the weights. The clothing area corresponding to each bone is found by identifying the body area with a weight greater than the threshold for each bone. For body points on a bone with a weight greater than the threshold... Find the point on the skeletal line segment that is away from the body. nearest point ,by Starting point Find clothing points along the direction, and record the point with the largest weighted centroid of the triangle containing the found clothing point as... All found Recorded in a collection middle; (2) Initial migration of clothing with corresponding skeletal structures: For a set In Find the distance on that bone nearest point and find the line segment Intersection with the body model ; The points on the transferred human body model and skeleton are obtained by utilizing the correspondence between the human body model before and after the transfer. and The new clothing point is calculated using the following formula. : （1） (3) Sampling multiple layers and closely spaced areas: Uniformly sample the vertices of the mesh on the clothing, and obtain a point after each sampling. , satisfy ,in This indicates the number of the sampling point. The sampling rate represents uniform sampling. This represents the total number of vertices in the clothing model; from Starting from a point, in its normal direction and the opposite direction of the normal, at a distance of threshold. All points obtained by the intersection of the inner part and its clothing are denoted as Construct its point-to-point sequence And record all the sequences obtained from sampling; (4) Initial migration of clothing points that do not have a corresponding skeletal point relationship: Using clothing points with corresponding skeletal relationships as constraints, and leveraging the continuity of the original clothing mesh, biharmonic interpolation is used to obtain preliminary migration results for clothing points without skeletal relationships. Simultaneously, the constraint step (3) ensures that the displacements of point pairs in the point pair sequence are equal, thus obtaining the preliminary skinning results for the migration of all points including unconstrained points. ; (5) Using clothing points with skeletal relationships as constraints, interpolate to obtain the local coordinate system corresponding to clothing points without skeletal relationships: Export the local coordinate system and coordinate axes of each bone in the human model before and after migration from the Autorig bone controller plugin. Simultaneously calculate the Euler angles from the original coordinate system to the local coordinate axes of each bone. The Euler angles corresponding to the human bones before migration are denoted as follows: The Euler angles of the human skeleton of the migration target are denoted as . Utilizing the continuity of the clothing mesh, biharmonic interpolation is used, with the Euler angles of the skeletal coordinate system corresponding to the clothing point as the boundary, to obtain the Euler angles of the local coordinate system corresponding to all points including unconstrained points; the transformation matrix that rotates the original coordinate system according to the Euler angles corresponding to the clothing point before migration is denoted as... ; The transformation matrix of the Euler angle rotation corresponding to the clothing point of the migration target is denoted as: ; (6) Transform the point pairs in step (3) into linear constraints in energy optimization: For a set of point pairs in the point pair sequence obtained in step (3) Construct the following formula (2): （2） in express Dot at In the results Coordinates, for and The same applies to coordinates; , Represents line segment Length, In practice, the relaxation constant for this location is taken as 0.
3. For all point pairs obtained in step (3) The coordinates are all constrained as shown in formula (3), and a linear constraint matrix is added. middle; (7) Construct the following energy function for each point on the garment: After obtaining preliminary migration skinning results Then, construct the triangular facets before and after optimization. The matrix representation of this matrix on the original garment is denoted as... ,definition ; in Let the column vector represent the coordinates of the three vertices of the triangle. ,in This represents the normal vector of the triangle; The same principle applies to the target clothing to be optimized. Matrix, recorded as ; The gradient of each triangle face transformation in the clothing migration is then expressed as: Then, the energy of each triangular facet is accumulated to obtain the conformal energy. Then update the energy components in sequence. (8) Eliminate clipping between clothing and the human body: The clipping of clothing and human body is divided into clipping between clothing points and human body models and clipping between clothing surfaces and human body models, and then these two types of states are judged respectively.
2. The multi-layer clothing transfer method preserving design semantics according to claim 1, characterized in that, In step (7), energy accumulation is used to obtain conformal energy. The mathematical expression is: （3） in Let Frobenius be the norm of the matrix. For an ideal transformation that only involves transformations within the triangular face, that is, in The transformation preserves the normal of the triangular facet; then the energy is updated in the following order: (7.1) Update the corresponding values of each facet in the current state. Matrix; First, project each transferred triangular facet onto the plane containing the transferred facet, and obtain The process is shown in equation (4) below: （4） Thus, the projected triangle is obtained. The matrix representation of is, i.e., When satisfied Update Otherwise keep Unchanged, making updates possible Energy decreases at this time; (7.2) Construction Construct shape-preserving energy according to formula (3); (7.3) Constructing energy to prevent torsion For clothing points corresponding to skeletal points in step (2), construct the energy as shown in equation (5): （5） in It is a constant, set as , , Represents the total points of the clothing. The total number of clothing points with skeletal markings in step (2); The vector representing the connection between the current optimization result and the skinning result obtained in step (4); Indicates the bone direction corresponding to the current point, using Indicates and Collinear unit vectors, then ;pass Can be kept exist While restricting directional freedom, it also limits movement to a plane orthogonal to prevent torsion. (7.4) For areas marked as needing to be closely attached The energy of the tightly attached portion is constructed as described in equation (6). : （6） (7.5) Change the current constraint terms Substitute the values and use the ADMM method to optimize while satisfying the R constraint. The quadratic programming problem is solved to obtain the final optimized clothing mesh result, which is recorded as follows: .
3. The multi-layer clothing transfer method preserving design semantics according to claim 2, characterized in that, Step (8) specifically includes the following sub-steps: (8.1) The first state is where there is clipping between clothing points and the human body model: For each point on the clothing, find the point on the human body model that is closest to this point, and at the same time determine whether this point is inside the human body model based on the normal direction of the triangle containing this point; all points inside the human body model And the nearest clothing store to this point. and the corresponding human face normal. ,exist Add constraints as shown in equation (7): （7） (8.2) The second state is when there is no clipping between the clothing point and the human body model, and an intersection between the human body model and the clothing model is detected: for each set of intersecting triangles, the points are... And the nearest clothing store to this point. And the normal direction of this human body. ,exist Add constraints as shown in equation (7); Repeat steps (7) and (8) until there are no new constraints in the last step (8) and the energy of step (7) converges, thus obtaining the result of eliminating clipping between the character and clothing.