Research On Reconstruction Technology Of Texture Images For 3D Human Model

At present,with the continuous development of computer science and technology,especially the visualization technology,a breakthrough has been made.The reconstruction of three-dimensional human body model has drawn extensive attention of researchers both at home and abroad,and has achieved some achievements,which has become one of the hot research topics in the fields of computer graphics.There is considerable practical value and broad applicaiton prospect in many fields,such as military training,aerospace,security monitoring and video games,for high quality 3D huamn models.However,due to the constraints of hardware conditions and some objective factors,there are still some difficult problems to be solved in the reconstruction of 3D human model,such as the texture seams caused by error and the texture images being not clear enough.Therefore,the fast reconstruction of 3D human model and generation of surface texture are regarded as the starting points,and the related problems are mainly studied in this thesis.The main research contents and innovations of this thesis are as follows:1.A 3D human body rapid acquisition and reconstruction system consisting of multiple cameras is designed and constructed.Through the full coverage by multiple cameras from corresponding perspectives,the consuming time of data acquisition process is about 1 second,which greatly reduces the acquisition time,and this can not be achieved in previous related work.After a series of processing,such as data optimization,multiple registration and surface reconstruction,the reconstructed 3D human body model has high precision,and the error is not more than 5mm.The experimental results demonstrate that the system can reconstruct high accuracy 3D human models on the basis of rapid data acquisition.2.A seamless images stitching algorithm based on adaptive iteration is proposed.In order to generate high quality texture images on the surface of the 3D human body model,an adaptive iterative solution strategy is proposed to solve the unmatched seams in the texture image.Then the texture image space can be expanded to move the texture image,thus the seams can be effectively eliminated from the texture images on the surface of the model.Finally,a seamless,consistent and high quality texture is generated on the surface of the 3D human model through the process of image fusion and restoration of the blank regions.The experimental results demostrate that the proposed algorithm can effectively eliminate the defects in the texture on the surface of model.3.Since the resolution of the color image taken by deep camera is low,the quality of texture images on the surface of 3D human body model is not good enough.A super-resolution reconstruction algorithm based on image segmentation,classification and sparse representation is proposed.As we know,there are different features in different kinds of images.Then the image is segmented into different regions,and classified into corresponding image categories.Finally the high-resolution image is reconstructed by the dictionary of each category based on sparse representation.The experimental results demonstrate that the proposed algorithm obtains better reconstruction performance both in visual effects and quantitative analysis,and the texture images reconstructed by our algorithm also get better quality.4.In order to improve the quality of texture images on the surface of 3D human models,a super-resolution reconstruction algorithm based on the Learned Iterative Shrinkage and Thresholding Algorithm(LISTA)at both ends with sparse representation is proposed.Two LISTA network structures are constructed to learn the high-resolution and low-resolution images.As the corresponding sparse codes are obtained,a single-layer convolutional neural network is constructed to study the nonlinear mapping between them.Finally,a complete deep learning network is constructed by connecting the above structures with each other.The experimental results demonstrate that the proposed algorithm obtains better reconstruction performance both in visual effects and quantitative analysis,and the texture images reconstructed by our algorithm also get better quality.