01-06-2012, 11:01 AM
2D-3D POSE INVARIANT FACE RECOGNITION SYSTEM FOR MULTIMEDIA APPLICATIONS
2D-3D POSE INVARIANT FACE RECOGNITION SYSTEM.pdf (Size: 424.91 KB / Downloads: 33)
Introduction: Face Recognition in Multimedia Applications
One of the reasons face recognition has attracted so much research attention and development over the past 30 years is its great potential in numerous multimedia commercial applications. Zhao and Chellappa [Zhao06] grouped face recognition technology into five different categories of multimedia applications and described their advantages and disadvantages. Another example of the evolution of face processing research is the continuously proposal of different benchmarks and per-formance evaluation initiatives [FRVT02, FRGC05], indicating that the field is far from maturity. In this subsection, face processing approaches are briefly described or mentioned depending on how suitable they are for the addressed multimedia application scenario. A total of five different scenarios have been proposed: Access Control Points, Intelligent Human Computer Interfaces, Virtual Collabora-tion, Video Indexing and Retrieval and Video Surveillance.
Mixed 2D-3D Face Recognition Schemes
Recently some of the new face recognition strategies tend to overcome the dif-ferent challenges from a 3D perspective. The 3D data points corresponding to the surface of the face may be acquired using different alternatives: a multi camera system (stereoscopy), structured light, range cameras or 3D laser and scanner de-vices. The main advantage of using 3D data is that depth information does not de-pend on pose and illumination and therefore the representation of the object do not change with these parameters, making the whole system more robust. However, the main drawback of the majority of 3D face recognition approaches is that they need all the elements of the system to be well calibrated and synchronized to ac-quire accurate 3D data (texture and depth maps). Moreover, most of them also re-quire the cooperation or collaboration of the subject making them not useful for uncontrolled or semi-controlled scenarios where the only input of the algorithms will be a 2D intensity image acquired from a single camera.
CONCLUSIONS
Although the experiments have been carried out on a small database, the idea is to present a framework for the extension of any statistical approach such as LDA, ICA, Kernel-PCA to this novel 2D-3D mixed framework. Thus, the performance of the Partial PCA may decrease for some scenarios such as Video Surveillance since PCA is sensitive to illumination changes. However, if there were enough training samples, a Partial LDA or even a Partial ICA could be implemented to make the face recognition system more robust towards these illumination changes.
Partial PCA is supposed to get good results for Access Control, HCI or VC sce-narios. In those scenarios, the main challenge is pose variation and as already demonstrated in the previous section P2CA is developed considering this chal-lenge. In these 3 scenarios the illumination may not vary considerably and the res-olution of the images is medium high. For the Video Surveillance it may depend a lot on the type of room or place of the scenario.