基于信息融合的人脸识别研究-计算机应用技术专业论文.docxVIP

优秀毕业论文 精品参考文献资料 研究生优秀毕业论文 Abstmct Abstmct 硕士学位论文 Abstract Based on也e high identiflcation of 2D facial texture features and the good robustness against variation of position，illumination，aJld expression of 3D face data，we focus on the geome仃ic feature ex仃action of 3D face poillt cloud，feature ex仃action of 3D depth image and 2D face image and如sion memods of face recognition．Our main work iIlcludes (1)While consid砸ng t11e panicul撕t)，of the po协t cloud of nose region，we presents a me也od for registration using the local poiIlt c10ud of face to correct face posture．Firstl y’me registration pe南mance alld e伍ciency of t11e ICP algor砒11：n are weighted to detemline the area size．Wim the symme仃y of face，t11e holes generated dudng data acquisition process can be filled by mirror． Experiment results show tllat it’s e丘’ectiVe to correct posture by nle aurea around nle nose tip when we weigh t11e regis仃ation error agaillst the costs of time． (2)Abom t11e point cloud geometric features，t11e most似o r印resentatiVe contour lines of face are ex仃acted．Conside血g nle robus廿less agaillst position and eXpression，nose aurea contours featIlres wi也out mou廿1 a11d face edge aure ex仃acted．Then identi毋t11e local contour lines t11at ex仃acted，aJld analyze nle证1pact of di丘’erent extemal factors．n shows the hi曲 discrimillation of ea’ective 10cal contouT features．Further more，recognition rate has been improVed on di疏rent test set when如sillg铆o contour features with di骶rent weights．The two contoulls had di虢rent innuence on recognition rate． (3)Multiple inf0肋ation如sion is studied i11 detail．On the one hand，the weighted LBP feature of 2D face images and 3D depm images are extracted．They are weighted to如se i11 matching layer to recognition．On t11e otller haIld，山e face contour features，2D face image f已atures and 3D d印t