Three Dimensional Human Face Acquisition for Recognition

Machine identification and recognition of human faces is a rapidly growing research area in both the academic and commercial world. Most of the research to date has concentrated on the use of two dimensional information, acquired from video cameras or photographs. The use of a three dimensional system is hoped to remove many of the problems affecting the two dimensional systems such as disruption caused by changes in the face’s orientation or changes in the ambient lighting. A three dimensional system will obviously not be influenced by orientation changes and the lighting is irrelevant, as it is the shape not the shading of the face that is important. For this system to be of practical use it is important that the process of acquiring the necessary information to generate the three dimensional surface model should not require any complex or expensive equipment and should not impose any undue constraints on the human subject, such as a requirement to remain still for an extended period of time. The described system uses a projected structured light, stereo vision arrangement to acquire the surface shape. Traditionally these systems have either been limited to simple surfaces or have required multiple views for more complex objects to prevent ambiguity problems in the stereo matching phase. This dissertation describes a novel method, loosely based on graph theory, to resolve this problem and thus requires only a single stereo image. The individual stages in the extraction of the surface are discussed and demonstrated: (i) the capture of the necessary stereo view (ii) the recovery of the features in the camera image (iii) the stereo matching of these features to the projected pattern using the new graph based technique, and (iv) the creation of the surface model. This dissertation concludes with an evaluation of the success of the system and suggests possible areas of further work including the steps needed to produce a successful recognition or identification system.