基于卷积神经网络的人脸检测和性别识别研究-计算机应用技术专业论文.docxVIP

研究生优秀毕业论文 Abstract Abstract 硕士学位论文 Abstract COnvolutional neural network is a special kind of deep neuraI network model，which is a new type of anificial neural ne帆orks a11d anificial neural netWork tecllllology combined with deep leaHling produced witll local experience area，hierarchy，feature extraction aIld classification仃aining process combined with global characteristics，in the field of image recogIlition has been widely applied．Panicularity renects conV01ution neural net、^，orks in似o waVs．on the one haIld itfs a non．coIlIlection be似een neurons mlly colulected，on the other haIld me s锄e Iayer some right coIulections between neurons weight is shared，this non—full comlectivi付and weights shared network architecture reduces the complexit)r of the ne帆ork model，reducing the n啪ber of weights，the network stmcmre of the tmslation，rotation，tilt， scaling aIld otherhighly inV碰ant． nlis anicle is designed to detecthighly v撕able caJl be robust face detection algorit№ based on conV01ution neural ne铆ork stmcture of face images automatically舶m a铲oup of people face aJld no face images in the training set，the automatic synmesis of specific issues simDle f已ature extraction，谢thout me need to ma：ke aIly assumptions or use of any mallual f-eature exⅡ孤tion aIld fIace area．ConV01ution neural netWork stmcture and training methods are desigIled for rotation，occlusion complex backgroulld image with robust be似een detection rate aIld false al锄rate achieved good balance．Experimental results show mat aJl efrectiVe face detect．on system before the image region classiflcation does not require any expenslVe local pretreatment．Thesis algorimms focus on nlree chalIenging to achieVe Very lli曲data rate of￡lce detection，paIticulally low false al锄rate，without the need to use muhiple netwDrks to haI