Automatic Animated Face Modeling Using Multiview Video

Abstract

An image-based 3-D modeling system is presented in this paper. Our modeling system consists of three main stages: camera calibration, depth estimation and 3-D geometry reconstruction. All of these steps are executed automatically. In the camera calibration stage, some patterns are used that help to determine the camera’s position in an environment. The camera’s intrinsic and external parameters are determined using epipolar geometry. After the camera parameters are determined, the camera’s location in each projected frame is determined. The depth for each pixel in a base image is estimated from the camera’s focus to the object’s surface by measuring the similarity between the base image and the neighboring images. The object’s 3-D geometry is reconstructed with texture from the base image using the depth information.