Research On Cross-age Face Recognition Technology

With the development of human-computer interaction technology,image intelligence application technology has received great attention recent years.As one of the hot technologies,face recognition has high research value and application prospects,which has been widely used in information security,civil air defense and other fields.However,face recognition technology is also facing enormous challenges due to factors such as age,expression and posture.As an important attribute of face,age has a significant impact on the reliability of results.Face aging is a complex nonlinear internal person change.And facial appearance has significant texture and shape change with age increasing.Besides,different people have similar change at same age time and same person has different change at different age time.Therefore,modeling face aging is a very complicated process.In order to address this problem,this paper earried out related research and proposed two models for cross-age face recognition task.The main work has two aspects as following.Firstly,this paper proposed a deep cross-age face verification model.Influenced by face aging,facial features change significantly with increase of age span.In order to reduce the impact of age span for face recognition,we propose a deep cross-age face verification model.Traditional methods use manual design features as discriminant features,which have disadvantages that higher dimensions but lower discriminability,and age information cannot be eliminated effectively.The deep cross-age face verification model uses both identification and verification signals,and it uses convolutional neural network to obtain high-level semantic features with age as an effective supervised signal.A series of experiments on public cross-age datasets show that this model can reduce the impact of age span and improve recognition accuracy effectively on cross-age face recognition task.Secondly,this paper proposed a joint multi-task convolutional neural network model for cross-age face recognition.One popular way for cross-age face recognition is modeling it as a general facial classification problem.However,such method usually faces a problem:how to extract identity-sensitive features that are age-insensitive and obtain much robust age-invariant features effectively.To this point,we proposed a joint multi-task convolutional neural network framework.The proposed framework contains two tasks:age classification task and identity recognition task.The former is used to learn age-sensitive features that are identity-insensitive,while the latter is used to learn identity-sensitive features that are age-insensitive.In order to enhance feature learning process of the two tasks,the proposed model introduces a regular item to constrain both tasks each other.Experiments show that the method can obtain much robust age-invariant features and mitigate the impact of large age spans for face recognition effectively.