3D Facial Recognition Based On Manifold Learning

With the rapid development of the computer and biomedical engineering technology,the use of biometric recognition is a preferred way in ID identification. As one of the mostnatural and friendly biometrics, automatic facial recognition has become widely researchand application. However, how to decrease the number of feature dimension is a keytechnology in facial recognition. Manifold learning is a nonlinear dimensionality reductionalgorithm, has been widely exploited in face recognition. These methods often assume thatmultiple samples per person (MSPP) for feature extraction during the training phase. Inmany practical face recognition, small samples and multi-source data are always appeared.This thesis is mainly focused on the problems of small samples and fused multi-source data.Main work of innovation is listed as follows:,1. The overview of the conception and basic procedure of3D face recognition, themethods of3D face recognition and the evaluation of subspace-based face recognition arealso concerned. The algorithms of face recognition based on manifold are introduced indetails.2. Build a3D face recognition framework based on subspace-based face recognition,and apply manifold learning-LOGMAP into3D facial depth image recognition.Experimental results on the3D real-time face database are presented to demonstrate theefficacy of the proposed approach.3. Analyze of problems face recognition based on manifold learning, for example ofsmall samples, mutil-information integration and given solutions.We propose complexdomain logarithmic map algorithm (C-LOGMAP) based on logarithmic map algorithm(LOGMAP) in this paper. Experimental results on the3D real-time face database and3DTexas face database are presented to demonstrate the efficacy of the proposed approach.