Multi-model AAM framework for face image modeling

Abstract

Active Appearance Modeling (AAM) offers acceptable face synthesis performance when applied to person-specific modeling applications. The aim of the work presented in this paper is to enable AAM to model and synthesize more accurately previously unseen face images. Thus a clustering process based on shape similarities is incorporated in the system and applied prior to conventional AAM modeling, to yield Multi-Model AAM. In this approach the wide appearance spectrum possible face images is decomposed into a number of cluster each containing similar shape faces. This allows AAM modeling per cluster to be applied and therefore the generation of several AAM models which capture more accurately variability between possible input faces. Experimental results show that, when dealing with previously unseen faces, models generated through this Multi-Model AAM framework can be significantly more effective in terms of both shape and texture, than the conventional single model AAM approach.