3D Face Recognition Based On Biomimetic Pattern Recognition

Three-dimensional face recognition is not only the cross-task of some subjects such as applied mathematics and pattern recognition, but also one of hot topics of biomimetic identification research. Three-dimensional face models possess richer information than those of two-dimension, which are less susceptible to the impact of illumination and posture, so three-dimensional face models are attracting more attention.In this paper, a new method of pattern recognition called biomimetic pattern recognition is applied into three-dimensional face recognition, using algorithms of super sausage neuron and triangular neuron in multi-weights neuron to recognize three-dimensional face features. Biomimetic pattern recognition is different from the traditional one. It is recognition from the "cognition" point of view, but not the "classification". And it is the analysis and "cognition" about the "shape" of high-dimensional complex geometrical shape which is formed by all the learning samples.The main work of this paper contains two parts:(1) Extracting features. First of all, Gaussian filter is used to reduce the noise of Range image and the areas of eye corner points and nose tip point are got. Then the eye corner points and nose tip point are located according to the Shape Index of each point in these areas. Finally, the distances between points are used as features.(2) Feature recognition. Firstly, super sausage neural network and triangular neuron network are constructed, wherein the super sausage neural network is constructed by minimum spanning tree which is constructed by all points in the training sample set and found out by Kruskal algorithm. Lastly, the two above mentioned neural networks are used to recognize features.The experimental results show that biomimetic pattern recognition can be applied into three-dimensional face recognition, and has a much higher recognition rate.