Research Of The Expression-Robust 3D Face Recognition Algorithm

In the past two decades, face recognition has been one of the most important and attractive research areas in computer vision, and has a broad future of applications in the fields of video surveillance, time attendance, airport surveillance and so on. A great deal of research prove that although traditional 2D face recognition has achieved great success in control environment, it still faces influences such as illumination, pose, makeup and facial expressions and so on. Compared with 2D face recognition, as 3D face recognition based on 3D face data is free from the disturbance of illumination and gestures,3D face recognition is considered to have a broad space for development. However, it faces many difficulties and challenges, such as expression change. This paper studies the advantages and development of 3D face recognition, and with regard to the problems such as expression, the two efficient solutions and appropriate recognition algorithms are given. The main work is listed below:1. A novel 3D face recognition algorithm based on facial profiles is proposed. First, the pose of a cropped face is automatically corrected based on principal component analysis(PCA), and all facial scans are transformed into the uniform pose coordinate system. Then, a set of facial profiles in the semi-rigid region are extracted to represent a 3D facial scan. As a result, the shape of two facial scans can be matched by matching the shape of their corresponding facial profiles. An open curve analysis algorithm is applied to calculate the geodesic distance between a pair of facial profiles extracted from different facial scans. The geodesic distance is used as a similarity measure. Finally, two facial scans can be matched using the weighted sum of all levels of their corresponding geodesic distance. Experiments carried out in the FRGC v2.0 dataset demonstrate the effectiveness of our algorithm, it is also robust to expression.2. A 3D face recognition method based on expression-invariant feature is proposed. First, every face is aligned to the generic face and is also registered with the generic face to achieve correspondences between two point clouds. Then, the face shape residue space and face expression residue space are constructed by the training set and the generic face, and shape residue between the input face and the generic face is projected to the two spaces to get face shape feature vector and expression feature vector. Finally, the features which are free from expression changes are extracted and used for recognition. Experimental results certify the effectiveness of our algorithm, and our algorithm effectively reduces the effect caused by expression changes.