AUTOMATIC DETECTION OF FACIAL FEATURES FOR 3D MODEL BASED CODING APPLICATIONS

Abstract

This dissertation presents an automatic detection scheme of facial features for 30 model based coding applications. This system can be used as a pre-requirement of a remote teaching application using 3D model based coding. A selected set of control points of the face is transmitted to the remote terminal instead of sending video signal. In order to extract this set of control points a predefined 3D generic wire frame model is used. In this project, feature points of facial image for 3D model fitting application were extracted. As the project intends to develop the automatic detection system, there is no manual detection for feature points. The first step is initial framework to the input image. The iterative processes of framework are low pass filtering, thresholding and edge detection. To detect the accurate feature points, it is needed to estimate the feature windows, which include the feature points. Secondly, the feature windows can be estimated by using vertical integral projection method. This method is also useful for detection of some feature points such as head top, chin points, eye center, mouth center and nose center. The third step is to extract the accurate center points of eyebrows. The centroid method is reasonably successful for eyebrow center detection. Last, four points of the mouth feature are detected with both Canny edge detection method and amplitude projection method. The first one had limited success and second gave a quite satisfactory result and in our final implementation we used the integral projection method. On the whole, the results obtained were encouraging and could be used for the initializing stage of a 3D model based coding system.