Font Size: a A A

Personalized Facial Modeling

Posted on:2009-12-15Degree:MasterType:Thesis
Country:ChinaCandidate:Z YueFull Text:PDF
GTID:2178360242983012Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Photorealistic facial modeling and animation is a research hot topic in computer animation, computer games, computer vision, and image recognition fields in recent years because of its wide applications in visual communication, computer games, etc. After reviewing the progress of facial modeling in computer graphics field in the past 30 years, we propose our solution. We try to model a person from a single photograph of front view. First, we capture a front view of a face from an ordinary camera. Second, we use the active shape model (ASM) as the feature detection tool to find the 77 features' location, which are based on the MPEG-4 standard. They are chosen according to following criterion: they can be seen from the front view, and they mark the features on face (eyes, nose, mouth ... and so on).In our current work, we use 65 of the 77 feature marks in our modeling process, and the other 13 marks will be used in future work. Third, we deform a generic geometry model, which is predefined as an average of Asian people from FaceGen, to the personalized model using the 65 features converted to 3D coordinates. Fourth, we make the texture from cutting out the face part from the photograph whose contour is shown by the feature marks, and fill the blank using the average color of cheek, and the boundary is blurred for smooth sake. Last, we compute the texture coordinate for each vertex on the deformed model (It is a simple affine transformation), and texture it on the model, which makes the model more realistic. In our project, we use the Laplacian editing as the mathematical tool in personalized model fitting stage, and compare with the Radial Basis Function (RBF) based interpolation method, which is used in the previous research. Our approach outperforms the previous ones in personalized facial modeling. The experiments show that our model fitting method can avoid part of the unwanted bad features which happens in RBF fitting. Our system can make a realistic personalized face model easily without any user interactions.
Keywords/Search Tags:Facial Modeling, Laplacian Editing, Radial Basis Function, Computer Graphics, Feature Detection, Active Shape Model
PDF Full Text Request
Related items