In this dissertation, we propose a distinctive facial feature for a pose invariant 3D face recognition method using Multiple Point Signature. We acquire 3D face models using the different devices and input faces which include different poses. All data...
In this dissertation, we propose a distinctive facial feature for a pose invariant 3D face recognition method using Multiple Point Signature. We acquire 3D face models using the different devices and input faces which include different poses. All data must be preprocessed and normalized using EC-SVD. We extract the invariant facial feature point using shape indexes and depth values from the range image. We propose a Multiple Point Signature method for measuring global facial surface information. A Multiple Point Signature represents several one-dimensional signatures. This method may not use error tolerance band to deal with noisy data and a registration technique to matching signatures. In addition, it is invariant to translation and rotation. Derivatives of facial surface are not required. Due to these characteristic, a Multiple Point Signature is fast and efficient to describe global information of facial surface. We use Euclidean distance matching for face recognition. We compared the proposed method with a single point signature method. From the experimental results, we have 94.6% recognition rate for the minimum distance matching, 85.4% by a single point signature. The proposed method shows 9.2% higher recognition rate than the single point signature.