人脸识别技术—计算机毕业设计(论文)

《人脸识别技术—计算机毕业设计(论文)》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、人脸识别算法TheDesignandImplementationofAlgorithmsforHumanFaceRecognition1-42人脸识别算法摘要人脸自动识别是模式识别领域的一项热门研究课题，有着十分广泛的应用前景。本文对人脸位置矫正，人脸的特征提取和识别这些方面进行了研究，并提出了相应的实现算法。人脸位置矫正作为人脸检测定位的一个环节，在计算机人脸识别中具有重要的意义。本文第二章提出了一种基于单人脸灰度图像中眼睛定位的人脸位置矫正方法，它是针对人眼灰度变化特点、人眼几何形状特征及双眼的轴对称性而设计的。实验结果表明，该方法对于双眼可见

2、单人脸灰度图像能实现快速有效矫正，并能在矫正结果中精确给出双眼瞳孔位置。本文第三章提出了一种基于神经网络的主元分析人脸图像识别方法。该方法利用非线性主元分析神经网络对人脸图像提取人脸特征（矢量），并在BP神经网络上实现了对人脸图像的识别。实验结果证明了该方法的有效性和稳定性。关键词人脸位置矫正，人脸特征提取，人脸识别，神经网络，灰度图像，图像块纵向复杂度，主元分析法，1-42TheDesignandImplementationofAlgorithmsforHumanFaceRecognitionStudent:YangboGuAdvisor:Dr.

3、WenmingCaoDepartmentofComputerScienceandTechnologyCollegeofInformationEngineeringZhejiangUniversityofTechnologyAbstractTheautomaticrecognitionofhumanfacesisahotspotinthefieldofpatternrecognition,whichhasawiderangeofpotentialapplications.Astheresultsofourin-depthresearch,twoalg

4、orithmsareproposed:oneforfaceposeadjustment,theotherforfacialfeatureextractionandfaceidentification.Faceposeadjustment,asaloopofhumanfacelocation,isveryimportantincomputerfacerecognition.Chapter2ofthisthesispresentsanewapproachtoautomaticfaceposeadjustmentongray-scalestaticima

5、geswithasingleface.Inafirststage,therightpositionsofeyesarepreciselydetectedaccordingtoseveraldesignedparameterswhichwellcharacterizethecomplexchangesofthegrayparameterinandaroundeyesandthegeometricalshapeofeyes.Duringthesecondstage,basedonthelocationandthesymmetryfeatureofeye

6、s,theinclinationangleiscalculatedandthefacepositionisredressed.Theexperimentationshowsthatthealgorithmperformsverywellbothintermsofrateandofefficiency.What’smore,duetothepreciselocationofeyes,theapplesoftheeyesaredetected.Inchapter3,anovelapproachtohumanfaceimagerecognitionbas

7、edonprincipalcomponentanalysisandneuralnetworkshasbeenproposed.ByusingBPneuralnetworks,humanfaceimagesaresuccessfullyclassifiedandrecognizedaccordingtotheoutputofBPNNwhoseinputistheeigenvectorextractedfromthehumanfaceimagesvianonlinearprincipalcomponentanalysisofasinglelayerne

8、uralnetwork.Simulationresultsdemonstratetheeffectivenessandst