A Multimodal and Multistage Face Recognition Method for Simulated Portrait

Recognition of simulated portrait obtained through face composition technique is a challenging task in public security area. An innovative method for simulated portrait recognition is presented in this paper. This method can be used to perform recognition of simulated portraits from anonymous cadaver, from the description of a witness, and from surveillance video. Based on principal component analysis (PCA), we have constructed eigenface, eigen-brow+eye, eigeneye, eigennose, and eigenmouth. Altogether 31 recognition modes can be formed through different weighted combinations of these eigen-parts. Face recognition of simulated portrait is first performed using this multimodal part based PCA (MMP-PCA) technique, results of which are then used as inputs for further recognition based on modified line segment Hausdorff distance (LHD). Experiment results show that this innovative recognition method has achieved good results for the recognition of simulated portraits in a database of 100,000 face images