Pose Computing And Recogniotion For3D Face Based On Geometrical Algebra Theory

Face data processing and recognition technology has been a research topic with practical application value and challenges, which will become the mainstream of biometrics in the21st century. In recent years,3D face data processing and recognition technology has opened up new avenues for3D face recognition research.3D face data carry geometry information of face surface, which are less effected when light, perspective, and scene change, and whose feature information is relatively robust. Since3D face expresses geometry information mostly by point cloud data, whose corresponding data processing technologies are different from the processing technologies of2D face data.According to the main line of3D face recognition, a non-classical algebraic—geometric algebra (GA) theory was put forward as geometric calculations and theoretical analysis method in this paper, and depth research was done from posture correction, posture estimation, feature extraction and recognition methods and other aspects of3D face. In this paper, geometric algebra operators rapid posture correction method of3D face posture correction and posture estimation method based on conformal geometric algebra relative geometric invariants are faster to achieve posture correction and estimation, based on which3D face sample data were standardized, including posture standardization, data structured and feature extraction. The calculation method and cover method of geometric constraints neuron model based on conformal geometric algebra. The main work and result of the text are as follows:1. According to3D face data of any freedom gesture, a correction method based on specific point location and geometric algebra rotation operator was proposed. The method views discrete Gaussian curvature value as a measure to determine the curvature of the critical areas on the face, in order to determine the front face, geometric algebra rotation operator is proposed for correction of operation. Experimental results show that this method is simple and geometric meaning is intuitive and accuracy is high.2. In line with3D face data of any freedom gesture, posture estimation method based on conformal geometric algebra relative invariant is proposed to determine6degrees of freedom geometric parameters of facial gestures, namely the translation parameters and rotational degrees of freedom. In this method, the geometric parameters of two relative geometric invariants are used to estimate the corresponding6degrees of freedom posture parameters, including the smallest unit of the ball and the smallest unit of axis from point clouds, and posture estimation is completed in the space of conformal geometric algebra. Experimental results show that the method is simple, and estimation accuracy is nice.3. In the light of these researches, samples were normalized to extract the features to prepare for recognition. In view of3D point cloud data samples of different postures, research strategy based on the thought of match in the space of geometric algebra is put forward. The method treats each standard sample posture as reference and calculates their relative geometry relationship, including rotation and translation, to determine the nose point to get a more point cloud data set with more standard pose, and by surface fitting and fixed-size template re-sampling to attain depth data, and extract corresponding feature by PCA.4. For3D face recognition, coverage model and calculation method based on conformal geometry space are proposed, including point constraint neuron model, line constraint neuron model and surface constraint neuron model. Finally, according to the thought of bionic pattern recognition, a line constraint neuron training algorithm and recognition algorithm based on CGA theory were proposed, and were applied to the3D face recognition experiments, and having the high generalization ability with small samples.The3D face calculation and recognition method based on GA new theory proposed in the paper are practicable and effective.3D face data generally have these features of many points, free space posture, rich expression and high feature dimension, and how to procee more effective applied geometric algebra, posture calculation and feature extraction and best coverage of space samples and recognition are research hotspots, and also the research direction for the future.