The Method Of Face Detection And Face Age And Gender Recognization

With the development of media and social networks,facial age and gender classification attracts extensive research interests due to their increasing numbers of applications.As biometrics identification based on face image has the advantage of non-contact,entertainment and convenience,it is widely applied in the fields of social networks,video surveillance,man-machine conversation and so on.In this paper,we mainly study face detection,face age and gender identification method,and we propose two solutions respectively,to adapt to different application scenarios.In the first scenario,the Faster R-CNN algorithm is adopted to detect face,and then we extract CNN features of faces to train and test.In the second scenario,we propose a face detection method,which is based on the ratio features and Adaboost algorithm,and then we extract the LBP features of faces.In the two above scenarios,random forest algorithm is proposed to train and test,after the features are extracted.The specific contents of the two schemes are as follows:(1)In the first scenario,as the Faster R-CNN algorithm has recently demonstrated impressive results on various object detection benchmarks,we train a Faster R-CNN model on the large scale WIDER face dataset,and evaluate the result on the FDDB face benchmark,the result shows that this method achieves high face detection rate.In order to improve the accuracy of face age and gender recognition in unconstrained environment,we propose a feature extraction method based on deep convolutional neural network,and we exploit a general-to-specific scheme.Firstly,we adopt VGG-Face model which is pre-trained on a large scale dataset for face identification.Secondly,we fine-tune VGG-Face on Celeb A with selected five attribute annotations and obtain face attribute model,the five attributes are as follows:In order to improve the accuracy of face age and gender recognition in unconstrained environment,we propose a feature extraction method based on deep convolutional neural network,and we exploit a general-to-specific scheme.Firstly,we adopt VGG-Face model which is pre-trained on a large scale dataset for face identification.Secondly,we fine-tune VGG-Face on CelebA with selected five attribute annotations and obtain face attribute model,the five attributes are as follows:?whether to keep a beard,?whether young,?whether to wear glasses,?gender is male or not,?whether to wear a hat.We use feature maps obtained at the fully connected layer as face feature vector.Finally,the random forest classifier is employed to train and test the Adience data set.Experimental results demonstrate that the CNN feature is robust,and the classification accuracy of the proposed method is high.(2)In the second scenario,a face detection method is proposed,which is based on the ratio features and Adaboost algorithm,and then we extract the LBP histogram features of faces.Specifically,a new image feature called ratio feature is proposed,it describes the ratio between any two points in the image.The new feature is scale invariant,bounded.We propose a deep quadratic tree to learn the optimal subset of the ratio features and their combinations,so we can locate face by the learned rules,a single soft-cascade classifier is used to handle the face detection.And then,the image segmentation method is applied to extract LBP histogram features at all levels,the random forest algorithm is proposed to train and test.Compared with the above classification method based on the CNN features,the accuracy of this method is low.