Human Body Shape Reconstruction From Binary Image Using Convolutional Neural Network

High-precision personalized three-dimensional(3D)human body is the data basis for many fields such as virtual try-on,customization,online garment retailing,and body health assessment or analysis,etc.There are mainly two methods to generate a 3D human body.The first method is to obtain the high-precision 3D human body through scanning the human body;the second method is generating the personalized 3D human body model by deforming the template of human body using data driving.However,as for generating a personalized 3D human body,the former method has the problems of complicated operation and time-consuming,while the latter method cannot generate a more accurate 3D human body.Therefore,we propose a new method for accurate reconstruction of the human body by inputting binary images from either single view or multiple views.We simplify the input and using the convolutional neural network to reconstruct the human body,which improves the problem which occurs beforeIn order to quickly reconstruct the three-dimensional human body through the minimum input,we first encode the shape of the human body via principal component analysis(PCA)or Auto Encoder(AE)to extract the low dimensional shape descriptor.Secondly,we extract binary images of different views of the human body.Then we design a novel Body Reconstruction Convolutional Neural Network(BRCNN)with two branches,which could capture deep correlated features from different views and merge them.Finally,we jointly train the BRCNN to learn a global mapping from the input to the shape descriptor which can be then decoded to points cloud for the reconstruction of various body shapes under neutral posesThe experimental results show that compared with the existing human reconstruction technology,the technology we proposed is one of the most convenient reconstruction algorithms There is no need for the customers to undress or scan themselves,the 3D human body can be reconstructed perfectly through just two photos taken by themselves.Further investigation also reveals that the accuracy has been improved and the prediction results of the two views are better than those from the single view,the prediction results of the weight-sharing network are better than the network without weight-sharing,the prediction results taken the binary masks as input is better than the results which take the sketches as input,and the prediction results taken PCA as human body encoded methods is better than the prediction results which taken AE as the encoded method of human body.