Hybrid Profile Recognition on the MUGSHOT Database

Face recognition is employed in several systems and computer based applications besides security applications. A broad variety of appearance based approaches has been presented and evaluated even on large datasets. However, most of the systems mainly concentrate on images and models that have been taken from the frontal view of the face. Thus they are limited to test views being taken from a similar viewpoint. In the present treatise we introduce several integrated hybrid systems consisting of artificial neural networks and statistical hidden Markov models to recognize profile views of unfamiliar persons, although the system has just been trained on frontal material. This problem is called the mugshot recognition task.