Research On Sampling In Manifold Space By Genetic Algorithm For3D Cartoon Modeling

3D Cartoon face is a famous computer model, which is based on original face photosfor3D modeling and artistic deformation. Compared with its original faces, it has bothsimilarity and exaggeration of art. Due to this, it gives people a better affinity and receivesan increasingly wide range of applications in many fields such as animated films, games,virtual reality and so on.The two indexes in cartoon face, namely similarity and art usuallyinfluence each other and is hard to control.The current researches of face similarity mainly focus to homogeneous face data, butfew researches have done on similarities among isomerous faces. At the same time, fewstudies have done on the face similarity model for guiding cartoon face synthesis.This dissertation makes study of the similarities between isomerous faces and theartistic features of3D cartoon faces by using manifold learning and principal componentanalysis (PCA), and then establishes similarity model of isomerous faces and the artisticmodel of face modeling. At last, the interaction of these two models results in a method forgenerating3D cartoon faces.The more details of works are as follows：Firstly, this dissertation builds three kinds of data sets, which contain1000pictures of2D faces,1000models of3D faces and many others of3D cartoon faces. Feature extractionand alignment are carried out for those date sets.Secondly, this dissertation proposes a method for consulting the similarity ofisomerous faces in manifold space. By using Two-layer Laplace manifold alignment, itbuilds a similarity model between2D face image and3D facial mesh. In the similaritymodel, if the2D face and3D face represent the same individual, their projection values inthe embedding are the same; if the2D face and3D face are the similarity, their protectionvalues should be similar.Finally, this dissertation proposes a method for generating3D cartoon faces based onthe similarity model. It uses principal component analysis for extracting the artistic featuresof the data set of3D cartoon faces, and then generates the3D cartoon faces under theinteraction of similarity model and artistic features.The validity of the similarity model is testified by experiments, and those3D cartoonfaces that generated by our method are much better than the traditional method. A systembased on the results of this work is developed.