| Relief,a popular sculpture form,has been widely used in many fields such as coins,architecture and craft life for centuries.There are various forms of relief.Common relief contents include characters,patterns,animals and plants,etc.According to the degree of compression of the space,the reliefs are divided into two types: high relief and low relief.This dissertation mainly studies the face bas-relief.The traditional portrait relief modeling relies heavily on the experience of craftsmen which is very time-consuming.With the development of computer science,a variety of automatically generated reliefs have been created in the field of relief modeling.Many great works have been proposed recently,but most methods have focused on the compression of model depth and preservation of details.In order to promote the development of automatic relief modeling and increase the degree of freedom of modeling,this dissertation proposes a method to relieve relief modeling limited to the input of 3D model pose.Combining face alignment and 3D deformation technology can automatically generate different expressions and poses bas-reliefs.The main work of this article is as follows:(1)In the alignment phase,a local face alignment method based on a 3DMM deformable model is proposed to ensure the quality of expression simulation.This dissertation uses PRNet to detect the facial feature points on the input picture and our 3D face model.Then we uses 3DMM to fit a 3D model with similar expressions to the input picture.Finally,both of the facial expression features of the fitted 3DMM model and the feature points of the input portrait image are considered to extract the facial expression features which would be applied to the input 3D model lately.The input face model is divided into several different regions so that different degrees of deformation can be adopted for different regions.(2)The deformation stage is divided into two parts.The first part uses the bi-Laplacian deformation algorithm to guide the expression simulation part,and the degrees of deformation is based on different regions;the second part use the skeleton-driven deformation algorithm.In this dissertation,we suppose the face model has two bones,and each vertex of the model is bound to the corresponding bones.Then,the pose of 3D model is adjusted to match the input picture.(3)Finally,the multi-pose and expressions bas-reliefs are generated by solving the sparse linear system in the normal field.In addition,the input 3D model is converted into a compressed height field as a height constraint in bas-relief modeling part.The resulting bas-relief retains the details of the original model while retaining the depth structure.Experimental results show that our method can effectively generate face portrait reliefs with different expressions and poses. |
