Research Of Facial Landmark Localization And Recognition For 3D Face

Face recognition is a kind of biometric recognition technology which identifies people based on their facial features. It has a broad perspective of applications in the fields of time attendance, access control, human-computer interaction and so on. Although 2D face recognition which is based on 2D images has achieved great success, it still faces difficulty in handling those problems caused by makeup, lighting, pose and so on. Compared with 2D face, 3D face model has more shape information which is free from the disturbance of lighting and pose. Hence, more and more researchers have been paying their attention to 3D face recognition in recent years. This thesis studies facial landmark localization and the recognition algorithm for 3D face which are two core issues of 3D face recognition. The main contributions of this thesis are as follows:1. Facial landmarks play an important role in applications such as processing and feature extraction for 3D face recognition. However, Facial expression and lighting have been the problem for the accuracy of landmarks localization all the time. In order to solve the problems, we propose an automatic and accurate facial landmark localization algorithm based on Active Shape Model and Gabor Wavelets Transformation, which can be applied to both 2D and 3D facial data. First, Active Shape Model is implemented to acquire landmarks' coarse areas. Then similarity maps are obtained by calculating the similarity between sets of Gabor jets at initial coarse positions and sets of Gabor bunches modeled by its corresponding manually marked landmarks in the training set. The point with the maximum value in each similarity map is extracted as its final facial landmark location. It is showed in our laboratory databases (OLD) and FRGC v2.0 that the algorithm could achieve accurate localization of facial landmarks with state-of-the-art accuracy and robustness.2. A novel method for 3D face representation based on Gabor and Log-Gabor texture distribution is proposed, and recognition experiments on OLD and FRGC v2.0 validate the effectiveness of the method. First, we discover that both the real and imaginary part distribution of Gabor and Log-Gabor can be modeled by generalized Gaussian density (GGD) when Gabor or Log-Gabor wavelet transformation (GWT or L-GWT) is applied to range face image. Second, the model coefficients of both Gabor and Log-Gabor are concatenate into a feature vector. Third, NLDA is applied to the feature vector for dimensionality reduction and 3D face recognition. The proposed feature is a relatively high-level feature in compared with the coefficients of Gabor and Log-Gabor Wavelet Transformation. It is more stable at the time of dimensionality reduction, and it enjoys the quality of both Gabor and Log-Gabor. Moreover, the proposed feature is robust to expression due to the fusion of the features on landmarks.