CN104376597B - A kind of hair method for reconstructing based on multi-direction constraint - Google Patents

Abstract

The invention discloses a hair reconstruction method based on multi-direction constraints, which reconstructs a complete hair model by using the hair surface geometry, hair surface direction and head model geometry reconstructed from images. The steps are: (1) based on the Laplace equation, using the hair surface direction to solve the hair spatial grid direction field; (2) based on the Laplace equation, using the hair surface geometry and the head model geometry to solve the hair spatial grid (3) According to the direction of the direction field in step (1), the gradient direction of the distance field in step (2) and the known hair surface direction, reconstruct the geometry of each hair from the evenly distributed hair root points of the scalp; ( 4) According to the initial geometry obtained in step (3), use the direction obtained in step (2) to optimize the initial hair model based on energy minimization to obtain the final hair model. The invention can reconstruct a hair model with even distribution of hair growth and similar appearance to the collected image.

Description

A hair reconstruction method based on multi-directional constraints

technical field

The invention belongs to the field of computer virtual reality, in particular, a hair reconstruction based on multi-directional constraints.

Background technique

In computer graphics, the correct modeling of virtual characters has always been an important topic for researchers. Whether it is special effects for movies, video games, virtual reality or other graphics-related fields, visually realistic character modeling technology has a wide range of applications. Hair is an important feature of a character. The shape of hair varies from person to person, and sometimes it is an important feature that distinguishes different people. At the same time, there are many strands of hair, and the shape, movement and optical properties of hair are very complex. Realistic hair modeling has become a research difficulty in computer graphics.

In the fields of film, animation or virtual reality, geometric modeling of 3D hair models has always been a tedious job. Currently, most models still rely on artists to manually shape and model hair using interactive tools. In recent years, researchers have been trying to use automated methods to reconstruct a model similar to real hair from image data. Image-based hair modeling methods generally obtain preliminary hair geometric constraints and direction constraints based on image information, and then reconstruct a complete hair model from them. At present, image-based 3D hair reconstruction methods generally first obtain the geometry of the hair surface, and then reconstruct the direction of the hair surface by analyzing the two-dimensional direction information of each image. Subsequently, one type of method directly spreads the surface direction field to the whole space, and uses the uniform direction field to reconstruct the hair, and the obtained hair is uniform enough but the shape is quite different from the image; the other type of method first reconstructs the outer hair, and then its Attached to the scalp, the resulting hair looks similar to the image, but does not guarantee uniform hair nor reconstruct the internal structure.

Compared with the traditional two methods, the present invention combines the advantages of the two methods, so that the hair grows in the uniform direction inside and in the surface direction outside, and combines the advantages of the two methods.

Therefore, the present invention's work on image-based hair modeling has great research significance and application prospects.

Contents of the invention

The technology of the present invention solves the problem: overcomes some limitations of the existing technology, provides a hair reconstruction method based on multi-directional constraints, effectively reconstructs the hair geometry of the hair after obtaining the preliminary hair surface data, and obtains a visually realistic hair model .

The technical solution of the present invention: take hair surface geometry, hair surface direction, and head model geometry as input, hair reconstruction based on multi-directional constraints, use gridded direction field and hair surface direction to jointly constrain hair reconstruction, so as to reconstruct hair Model. It is characterized by the following steps:

(1) Reconstruct the spatially gridded hair direction field from the input hair surface direction;

(2) Reconstruct a spatially gridded distance field from the input hair surface geometry and head model geometry;

(3) Using (1) the obtained direction field, the input hair surface direction and (2) the obtained distance field to reconstruct the hairline geometry;

(4) Utilize (1) gained direction field, optimize the hair model of (3) gained;

The advantages of the present invention are:

1. In the reconstruction process of the hair strands designed by the present invention, the strands of hair evenly grow from the scalp and gradually grow to the surface of the hair, the distribution of the strands of hair is even and reasonable, and the growth direction of the strands of hair is natural under the constraint of the direction of the spatial grid.

2. In the hair reconstruction process designed by the present invention, when the hair grows close to the scalp, it will grow according to the direction of the hair surface, and the outside of the hair is close to the collected image.

3. The multi-directional constrained hair reconstruction designed by the present invention makes the hair growth evenly distributed while maintaining the realistic external shape, which is convenient for the reuse of the reconstruction model.

Description of drawings

Fig. 1 is a data flow chart of the method of the present invention.

detailed description

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments.

The main flow chart of the inventive method is as shown in Figure 1, possessing hair surface geometry, hair surface direction, under head model geometry is input, concrete steps are as follows:

(1) Solve the hair direction field with spatial meshing.

The invention solves the spatial gridded hair direction field, which provides a continuously distributed grid direction field for the entire reconstruction area. First, according to the input hair surface geometry and the points in the head model geometry, the bounding box of the entire area is calculated, and a certain grid unit size is set, and the entire area is evenly divided into cubic grids to realize spatial gridding. Then, the Laplace equation is established, the direction of the grid containing the hair surface geometry in the grid is set as the direction of the hair geometry, the boundary grid of the bounding box is set as the Dirichlet boundary, and the other grids are unknown directions. Finally, the Laplace equation is solved to obtain a spatially smooth grid direction field.

(2) Spatial gridded distance field;

In order to make hair grow closer to the hair surface gradually from the scalp, the present invention uses the distance field to provide relative direction constraints. First, use the grid obtained by gridding in the step (1) to establish the Laplace equation in space, and the grid distance value containing the hair surface geometry in the grid is set to 0, including the head model geometry The grid distance value of the surface is set to -1, and the grid of the spatial bounding box boundary is set to 1. Then, the Laplace equation is solved to obtain the hair surface distance field continuous in space, where the hair surface is 0, the hair interior is negative, and the hair exterior is positive.

(3) Based on multi-directional hairline reconstruction.

This step is the core step of the present invention, starting from the scalp to rebuild each strand of hair. First, the user specifies the hair growth area on the surface of the head model, and evenly samples the root point in the area. The growth direction of the hair root is set to be between the direction of the head model and the tangent direction away from the hair parting line. Proportion. Then start to reconstruct the whole hair point by point, use ξ to represent a hair, ξ is composed of points ξ(0), ξ(1).....ξ(N) on it, the hair reconstruction starts from Starting from hair root ξ(0), it is reconstructed point by point, and is reconstructed for each ξ(i), which is determined by formula (1).

ξ(i)=ξ(i-1)+α ₁ V(ξ(i-1))+α ₂ ▽(D(ξ(i-1)))+α ₃ Ο(ξ(i-1)) +α ₄ S(ξ(i-1))(1)

Among them, ξ(i-1) is the position before ξ(i), V(ξ(i-1)) is the direction field value of (1) at the position of ξ(i-1), D(ξ (i-1)) is the distance field value of (2) at the position of ξ(i-1), ▽(D(ξ(i-1))) is the distance gradient direction of the position, Ο(ξ( i-1)) is the growth direction of ξ(i-1), the root node direction is determined by the above, and the rest is determined by ξ(i-1)-ξ(i-2), S(ξ(i-1)) is the geometric direction of the hair surface closest to ξ(i-1). In the formula, α ₁ , α ₂ , α ₃ , and α ₄ are the weights of constraints in each direction. The closer α ₁ and α ₂ are to the scalp, the greater the weight, and the closer α ₄ is to the hair surface, the greater the weight. α ₃ is a constant.

For each hair, when the distance field value D(ξ(i)) of its new reconstruction point ξ(i) is greater than a certain threshold, or the total length of the hair is greater than a certain threshold, the hair reconstruction ends.

(4) Hair optimization based on energy minimization.

The hair optimization based on energy minimization optimizes all the hair strands reconstructed in the step (3). Optimization is achieved by minimizing the energy E, which is determined by equation (2).

Among them, ξ ⁰ (i) is the initial hairline point reconstructed in the step (3), ξ(i) is the point to be sought, and the meanings of the directions of V and Ο are the same as those in the step (3), that is, V(ξ ⁰ (ii)) is the direction field value of (1) where ξ ⁰ (ii) is located, and Ο(ξ ⁰ (ii)) is the growth direction of ξ ⁰ (i-1). N(ξ ⁰ (i)) is the point set of the closest point to ξ ⁰ (i) on each hairline that is close enough to ξ ⁰ (i). w(ξ ⁰ _n (i _n )) is the Gaussian weight calculated according to the distance, and α _reg , α _ort , α _n are the coefficients of each energy item, all of which are constants.

The contents not described in detail in the description of the present invention belong to the prior art known to those skilled in the art.

The foregoing is only a preferred embodiment of the present invention, and it should be pointed out that for those skilled in the art, some improvements and modifications can also be made without departing from the principle of the present invention. It should be regarded as the protection scope of the present invention.

Claims (3)

1. a kind of hair method for reconstructing based on multi-direction constraint, its input is hair surface geometry, the hair surface field of direction and Head model, it is characterised in that comprise the following steps：

1) the spatial gridding hair field of direction is solved；

2) spatial gridding distance field is solved；

3) rebuild based on multidirectional hair；

4) Hair model based on energy minimization optimizes；

The step 1) in the spatial gridding hair field of direction solve, its step is：It is spatial discretization first, according to input Hair geometry and head model, calculate the spatial surrounding box of rebuild 3D region, and press Unified Element by space uniform discretization For grid；Secondly, the field of direction that Laplace's equation solves hair gridding, the grid side respectively comprising hair surface direction are set up To the surface direction is set to, the boundary condition of space lattice is set to Di Li Crays border；Finally, by solving the Laplce side Journey obtains the direction of each locality grid, so as to obtain the hair field of direction；

The step 3) in, to be rebuild based on multidirectional hair, its step is：First, user specifies natural on-off cycles of hair growth area in scalp Domain；Secondly, uniformly in growth district sampling root of hair point, the initial growth direction of per takes the normal direction by a certain percentage and cuts To between；Subsequently, using the multiple each hair of direction constraint reestablishing, hair is rebuild from the pointwise of root of hair point position, and one new to send out Direction of the position of silk point by former point, grid direction field direction, grid distance field gradient direction are nearest apart from former point Hair surface direction constrains determination jointly, and this process is realized using equation below：

$\mathrm{\ξ}\left(i\right)=\mathrm{\ξ}(i-1)+{\mathrm{\α}}_{1}V\left(\mathrm{\ξ}\right(i-1\left)\right)+{\mathrm{\α}}_2\▿\left(D\right(\mathrm{\ξ}\left(i-1\right)\left)\right)+{\mathrm{\α}}_{3}O\left(\mathrm{\ξ}\right(i-1\left)\right)+{\mathrm{\α}}_{4}S\left(\mathrm{\ξ}\right(i-1\left)\right)$

α in formula₁, α₂, α₃, α₄For the weight of all directions constraint, α₁, α₂Bigger, the α apart from the nearlyer weight of scalp₄Apart from hair surface Nearlyer weight is bigger, α₃For constant, ξ represents a hair, and ξ is made up of each point ξ (0) thereon, ξ (1) ... .. ξ (N), and ξ (i) is i.e. I-th point, ξ (i-1) the i.e. positions of ξ (i) former points, V (ξ (i-1)) are the step 1 of ξ (i-1) position) direction Value, D (ξ (i-1)) is the step 2 of ξ (i-1) position) distance field value,I.e. the position away from From gradient direction, Ο (ξ (i-1)) is the direction of growth of ξ (i-1), and S (ξ (i-1)) is several apart from the nearest hair surfaces of ξ (i-1) What direction；Finally, when each hair grows to maximum length, or the distance value of new hair point stops more than certain threshold value hair Growth.

2. a kind of hair method for reconstructing based on multi-direction constraint according to claim 1, it is characterised in that：The step 2) in, spatial gridding distance field is solved, and its step is as follows：First, using the step 1) in identical discretization grid build Vertical Laplace's equation, the grid distance comprising hair surface geometry are set to 0, and the grid distance comprising head surface is set to -1, net The distance on lattice border is set to 1.

3. a kind of hair method for reconstructing based on multi-direction constraint according to claim 1, it is characterised in that：The step 4) in, the Hair model based on energy minimization optimizes, and its key step is as follows：With the step 3) each hair point for reconstructing It is optimized for initial position, is constrained to suffered by each point：Various point locations and the initial position are close to, at the beginning of each point direction and the point The grid direction field value of beginning position is close to, the weighting direction of each point direction and the initial position neighborhood other hair closest approaches connects Closely.