The Key Technology In3D Face Recognition

Today, biometric face recognition is an important research content, and its recognition ismainly reflected in2D image field. Degree of recognition accuracy due to posture,illumination, expressions, movements and other changes, there are substantial limitations. Upto now, how to build a robust face recognition system is still a very complex difficult issue.However,3D facial model have more abundant information than2D face images, hence theuse of3D information of human face recognition is a more effective way.The thesis mainly focuses on3D face recognition problem. On the basis of the3D facemodels feature point extracting, to choose the feature points of the face facial rigid area, toweaken expressions factors, through the combining of coarse registration and fine registrationto implement registration, to transform face models into the consistent state, finally througheffective recognition algorithms implement recognition. The main job of the thesis issummarized as follows:In the part of “3D face feature points location”, radial difference image and Gaussianimage is used to detect the edge of face features. Through constructing radial differenceimage and Gaussian image, Region growth is proposed to construct the facial feature. Thealgorithms of fitting region edge and optimizing energy function are presented to extract theaccurate feature points, and providing a manual way to adjust the feature points.In the part of “3D face registration”, proposes a method based on the face facial rigid area,to locate feature points. In this method, the rigid area is not easily affected by the influence ofexpressions, can effectively achieve the registration. Improved ICP algorithm by using barrels,improved nearest point algorithm for searching speed.In the part of “3d face recognition”, existing recognition algorithms are carefullyreviewed and discussed in theory. Methods PCA, LDA, ICP and Hausdorff distance areimplemented, and their effects and accuracy are exhibited and analyzed with experimentresults. And proposes an improved ICP algorithm, by reducing iterations, effectivelyimproves the speed of recognition.