Comparing CBR and NN Performance in a 3D Face Recognition Application

摘要

The 3D face verification is a common task for humans, however for computers this is harder and slower. This paper shows results of two important paradigms in Artificial Intelligence -AI, Artificial Neural Networks -ANN and Case Based Reasoning -CBR, applied to 3D facial pattern classification. To test both approaches, 52 different individuals with their own variations in different spatial positions and expressions compound the 3D face database. In the Case Based Reasoning approach, the Hamming Distance evaluate the degree of similitude between cases. To improve the results a spatial normalization and a facial weight criterion are applied. In the Neural Networks approach by Radial Basis Function, the property of interpolation between faces and their variation, and the diversity of faces help to minimize the output error. The obtained results show that ANN's paradigm classifies better the faces patterns than the CBR paradigm, reaching zero False Acceptance Ratio -FAR and False Rejection Rate -FRR errors for the 3D face set used.