The Method Of Three Dimension Facial Model Reconstruction Based On The Numerous Photos

Three-dimensional model of the face can be widely used in theapplication of VR (Virtual Reality),Remedial Teaching,RemoteMeetings,HCI (Human Computer Interface),games and entertainment,film production,and so on. As individual differences of facial features is anobjective existence that is manifested in different texture features andgeometric features,to generate three-dimensional face specific model is ameaningful job. At present,the three-dimensional face model work in theforeign countries has become a hot topic,and in China there are someresearchers'studies in this area,but testing and practical experience showsthat the most 3D model of the face generated can not be satisfied and thealgorithm is too complex to be used.The 3D facial reconstruction based on the numerous photos is not onlya key issue in the face recognition research,but also one of the basicproblems in the field of computer vision and graphics. The key technologyis how to make the facial features calibration point automatic,in order toobtain more accurate information of two-dimensional feature points,andthe next one is to make two-dimensional with three-dimensional mapping,and last one is correcting model to get specific 3D face.Based on the front and side photos of face,a method of generating 3Dspecific facial model is proposed. Using binocular stereo vision methods,two-dimensional can be mapped to three-dimensional space,and theinitiative to shape models feature automatic calibration algorithm isimproved. QRPM algorithm is proposed to solve the problem of originalalgorithm search feature points trapping local minimum. Then the 3Dinformation of CANDIDE can be corrected according to the data,andusing B-spline interpolation to make the model be smooth. At last skin texture will be mapped on the face model,and thus 3D Face Model will bereconstructed with all characteristics.First of all,the mathematical model of two-dimensional informationmapping on the three-dimensional space is established,which is based onthe principle of binocular vision. Three-dimensional visual observes adifferent perspective of the number of 3D images; from the viewpoint overthe same scene,using triangulation principle,a three-dimensionalcomputer reconstruction of the shape and location of the binocular visionand depth have access to information. To reduce the impact of cameraexternal parameters and internal parameters and two-dimensionalinformation mapping on the three-dimensional space mathematical model isestablished to determine the spatial coordinates of the object in the samepoint on the image plane as the corresponding relationship between thepoints. With semi-automatic method,facial feature points will be marked.In accordance with mapping mathematical model,the experiment receivesa good result,and at the same time the experiment which does not have theanticipated effect will be analyzed,and gives solutions.Secondly,QRPM algorithm is proposed which is based on ASM. Sofar in the recent three-dimensional model system,the feature pointspositioning technology is not satisfied. In the field of face detectionresearch,researchers often only care about the human face probablylocation information,but not the characteristics of the accuracy of thepositioning point. And the majority of the reconstruction algorithm researchdone in the model experiments assumes that the feature location of the pointis given,for example,the position of the two eyes. However,the methodof positioning feature points is far from being a mature technology; theaccuracy of positioning feature point greatly affects the whole 3D Facialreconstruction system performance. If the deviation of positioning featurepoints is very great,it will lead to facial malformations phenomenon. Facial feature extraction is a very challenging work; the reason is that thegray facial little step,even for their eyes or mouth. And the general edgedetection algorithm is often difficult work. To take the initiative to shapefacial features model calibration method based on the model is popular inrecent years the general shape of object extraction algorithm,the core ideais a partial model matching points on the basis of the statistical model usingspecific treatment face constraints shape,which translated into anoptimization problem,and hopes that eventually converge to the actualshape of the human face forward. Its definition of an energy function,adjusts the model parameters to minimize the energy function,andaccording to the training data for the parameters to limit regulation,whichwill change the shape of a reasonable restriction within the scope. ActiveShape Model for classical algorithm model in facial feature points,sometimes does not be consistent with the emerging shape of the face oforgan distribution of the situation. To pursue the initial model is due toimproperly position,which led the search process in a local minimum. Ifthe initial position calibration model is given in the face near the mainorgan,or face regional focus and coordinate coincidence,the model willsearch the successful facial feature points. QRPM is proposed as theoptimization algorithm,which gives the initial location of the calibrationmodel,and improves strategies to avoid the emergence of illegal shape. AsASM has a good adaptive capacity,if a better initial position is provided,through multiple iteration it can find regional feature points. At the sametime the model of each sample point gray information is given,it can get asimilar coefficient to measure gray image of a region,and its similaritybetween the models,the degree of similarity can be obtained in the largestregional. So the initial position will be will be the focus of the algorithm.Experimental results show that the accuracy and robustness of improvedASM methodology are greatly enhanced. Finally,in order to build a visual model of the three-dimensionalhuman face,CANDIDE model features 3-D coordinates should becorrected. Generic model for a certain amount of feature points overall andpartial adjustments will be selected. And B-Spline interpolation will be usedfor smoothing texture,a specific 3D face model will be obtained.To sum up,some superficial research is carried out in the 3D Facialreconstruction. Amethod based on numerous photos of a face reconstructionis proposed,and the initiative to shape model for calibration feature theoriginal algorithm has been optimized and improved. QRPM algorithm isproposed to avoid local minimum. With scientific and technologicaldevelopment and social progress,and innovation of 3D FacialReconstruction model will continue. I hope that the method presented in thispaper can make a modest contribution to the development to the 3Dreconstruction.