Research On Face Recognition Based On 3D Model

Face recognition is one of the most hot and challengeable technologies, which based on biometrics, and also one of the most potential technologies. As the most natural and friendly way, automatic face recognition has become the important part of the Next Generation Computing. 3D face recognition methods are in general able to overcome the problems resulting from illumination, expression and pose variations in 2D face recognition. This thesis addresses itself to the task of 3D face recognition. The main contributions of the work are as follows:(1) A new triahglulation method based on profile is proposed, which is used to standardize the data generated from diversified methods. And a data struct of triangle index is designed. The storage quantity is reduced much, and the convenience of the manipulation is improved.(2) Radial difference image and improved Gaussian image are used to detect the edge of face features. Region growth is proposed to construct the facial feature. The algorithms of fitting region edge and optimizing energy function are presented to extract the accurate feature points.(3) The curve distance of two key feature points is added into the feature vector, which consists of Euclidean distance, curve distance, angle and volume.(4) The vertical direction of face is determined by the PCA at the random pose. Based on surface registration, the symmetrical surface and symmetrical profile are extracted. Curvatures on the symmetrical profile are calculated to extract the other three transverse profiles. At last, the four profiles are resampled and standardized. Only the valuable parts are used as the input of ICP algorithm.(5) MEGI model with the position information of each surface is proposed. Spherical correlation coefficient is extended to recognize the 3D face with MEGI model.(6) Some manifold algrithms are researched. A 3D face frame based on the manifold is presented.