Research On Portrait Relief Generation And 3D Portrait Reconstruction

Traditional relief design is completed by hand,not only time-consuming and laborious,but also requires the carver to have professional experience,which is difficult for ordinary users to complete.In recent years,some algorithm-based automated or semi-automatic relief modeling technology has been widely used,which effectively reduces the manual interaction and improves the modeling efficiency.However,it is still unable to achieve mass production of relief modeling,which can not meet the requirements of modern industry.Deep learning has gradually been applied in the field of relief modeling,However,a number of methods are suitable for general relief modeling design and there are few studies on portrait relief modeling.This article focuses on the portrait relief model as the research object,and study the two aspects of the forward compression of the 3D relief model to generate 2.5D bas-relief and from 2.5D bas-relief to 3D portrait reconstruction.Which include main aspects:(1)Portrait relief generation,a neural network framework suitable for portrait relief height field training is proposed.the input is a single-channel portrait relief height field,and the output is the compressed portrait relief height field.A convolutional neural network is established to train the height field of portrait relief.The training data set of portrait relief is constructed,and the detailed characteristics of portrait relief are learned through network training,which can eliminate the discontinuity problem of height field,and achieve high efficiency,low human-computer interaction and large volume portrait relief modeling.Due to the modeling efficiency of the method in this paper is much higher than the previous numerical optimization methods,this part also studies the generation technology of intelligent portrait relief animation,rendering the predicted portrait relief model one by one to make relief animation.(2)3D portrait reverse reconstruction technology,a single 2.5D portrait bas-relief reconstruction method of 3D portrait relief model,to provide more viewing angles for observers.Taking a bas-relief portrait model as the input,the 3D face reconstruction was firstly carried out,and then the whole 3D portrait reconstruction was completed.The key technologies used in the reconstruction included normal transfer and shape optimization based on template portrait.And according to the reconstruction of the 3D portrait model as the input,it can regenerate new portrait relief models of different styles and thicknesses.The newly generated portrait relief has a reasonable depth sorting and a similar appearance to the input model,which greatly improves the editability of portrait relief.