Researches On Identity Authentication Based On Eye Movement Spatiotem Poral Features

Identity authentication is a kind of security technology used to confirm the user’s identity by checking the identity items.In recent years,traditional identity authentication methods such as keys and certificates have been gradually replaced by biometric identification due to the disadvantages of being easily lost and inconvenient to carry.However,some of the current biometrics have some problems such as being easily forged or difficult to integrate with other biometrics.Eye movement is a kind of behavioral biometric that can reflect the activity of brain regions and eye muscles.It is not easy to be forged and has high security.In addition,eye movement information is located on the face so that it can be easily combined with other facial biometrics such as the human face and iris.Therefore,this thesis chooses eye movement to achieve identity authentication.First,by calculating the changes in the user’s concentration during the process of eye movement data collection,this thesis explores the rationality of eye movement recording duration;Then,by comparing the advantages and disadvantages of current features,a closed-set identity authentication method based on eye movement motion features is proposed.Finally,by analyzing of existing eye movement spatial features and referring to the application of metric learning in face recognition,an open-set authentication method based on spatiotemporal features is proposed.The main innovations of this thesis are as follows:1.Exploring a more reasonable duration of eye movement recording.First,because the eye movement recording time is too long and inconsistent in the existing eye movement-based authentication methods,the deviation angle of the gaze point relative to the screen stimulus point is used to represent the user’s concentration of visual stimuli during the eye movement data recording process.Second,on the fixation-only eye movement dataset,by comparing the deviation angle with the corresponding angle of foveal vision,whether the user maintains sufficient attention to the eye movement stimulus material is evaluated.Finally,a more reasonable eye movement recording duration is determined through the experimental results.On this basis,a dataset with shorter eye movement recording duration is constructed based on the existing dataset.2.A closed-set authentication method based on deep eye movement features is proposed.In order to obtain better authentication results on the dataset constructed,this thesis uses the spatial movement distance and the orientation of the fixation points to represent the motion information as eye movement motion feature,in order to overcome the limitation that saccades and fixations are processed separately in the most of existing methods.In addition,a deep network is designed for feature extraction and classification.The experimental results on the dataset constructed in this thesis show that this method achieves better closed-set authentication results than other methods.3.An open-set identity authentication method based on eye movement spatiotemporal features is proposed.Aiming to solve the problem that closed-set authentication can’t be trained on a large scale and can’t handle the problem that the test set identity category is larger than the training set,the following three steps are implemented: First,metric learning is introduced to achieve open-set classification.Second,the eye movement recordings are classified into fixation and saccade through the eye movement behavior classification algorithm.All saccades in each eye movement record are mapped into a feature map as the representation of identity.Finally,a metric learning network is designed to extract the spatial distribution features from the feature map.The spatial distribution features is fused with the eye movement motion features introduced in the previous section and performed to obtain the eye movement spatiotemporal features.The authentication results are obtained through the metric learning with multi-similarity loss.The experimental results show that the proposed method has improved the performance of open-set authentication compared with the comparison method,which shows the effectiveness of the proposed method.