MODEL-BASED 3-D FACE ANIMATION SYSTEM (LIP-SYNCHRONIZED) DESIGN FROM A VIDEO SOURCE
This thesis proposes a model-based 3-D talking head animation system and then constructs a simple 3-D face model and its animation by using Virtual Reality Modeling Language (VRML) 2.0 in conjunction with a VRML's Application Programming Interface (API), JAVA. The system extracts facial feature information from a digital video source. The face detection and facial feature extraction are prerequisite stages to track the key facial features throughout the video sequence. Face detection is done by using relevant facial information contained in the normalized YCbCr color space. Independent Component Analysis (ICA) approach is applied to the localized facial images to identify major facial components of a face. Then, an image processing approach is deployed to extract and track the key facial features precisely. Streams of the extracted and determined facial feature parameters are transferred to the animation control points of the designed VRML 3-D facial model. Since the face model is defined in the 3-D space while a given video source is a 2-D presentation, some heuristic rules are embedded to estimate the coordinates of unmeasurable points for the visually acceptable 3-D talking head model and animation. A standard video data set Miss America, QCIF formatted 30Hz, video is used as a test sequence, and the capability of the proposed system is verified and demonstrated.