A Hair Reconstruction Parameterized Model Via Volumetric Vector Fields

3D hair modeling plays an important role in the application of studio entertainment,3D gaming and so on.There're two types of hair modeling methods at present: manual editing using professional software for 3D modeling design;image-based modeling(IBM).The former requires the user being skillful in software operation and having a good sense of art,it is usually time consuming and laborious.The latter takes a portrait as input and finds a model from a 3D hair database that is most compatible to the portrait,which is highly efficient but lack of controllability.This paper investigates the parameterized representations of 3D hair data with which a user can create new 3D hair models by altering the value of parameters.Parameterized representations are used to capture shape and pose variation of human body,as well as shape and expression change of human face.The variety of hairstyles and complexity of hair geometries make establishing parameterized representations of hair models very challenging.In fact,we did not find related study on this issue.As the first attempt,we propose a parameterized representation for 3D hair models by analyzing node positions of hairlines.We obtain a 3D hair database including 343 hair models and each model consists of 10-thousand hairlines and each hairline contains 100 nodes.We concatenate all coordinates of all nodes of a hair model into a high-dimension vector and then perform principal component analysis(PCA)on such 343 vectors to extract a set of PCA components.A parameterized representation of hair models is then expressed as the sum of the linear combination of PCA components and the average of hair models.We then use the parameterized model to fit all original hair models in the database and determine the number of PCA components to be used in the representation according to the fitting error.In addition,we propose the other parameterized hair representation based on volumetric vector fields.An AABB bounding box is firstly calculated which contains all 3D hair models of the database inside.Then,we uniformly partition the box into a set of cells.Following,we create a volumetric vector field for each hair model with each cell filled a vector which is the average of the directions of hair strands passing through this cell,strands length and local density data is also recorded in the process.In addition,we extract Euler angles from the difference between each field and the average field.Finally,we perform sparse localized decomposition on these datasets to extract sparse localized components(SPLOC).And the parameterized models are then also expressed as the sum of the average and the linear combination of these components.We conduct a variety of experiments to compare the proposed parameterized models and analyze their advantages and disadvantages.Finally,with the proposed parameterized 3D hair representation,we design an interactive system for 3D hairstyle editing.Experimental results show that our system can create 3D hair models with a large variety of hairstyles by simply changing the value of blending weights of SPLOCs.