Cost-sensitive Sequential Face Recognition Based On Deep Active Learning

Face recognition technology is an important technology in the field of artificial intelligence and image information processing.This technology is designed to perform identity authentication tasks based on human facial features.Because of its non-invasive,concealed and support for large-scale identification,it is widely used in the military,border defense,justice,finance,education,medical and other fields.Since its inception in the 1960s,it has been one of the research hotspots and difficulties in the field of artificial intelligence and image information processing.This paper proposes a face recognition method based on cost-sensitive sequential three-way decision based on active learning and deep learning,in view of the high training cost of face recognition system and the pursuit of minimizing classification error but neglecting the misclassification cost imbalance.Firstly,the active learning algorithm is combined with the deep convolutional neural network to explore the dynamic incremental face recognition method in the case of sample mark scarcity,and the verification experiments are carried out on the FERET,CMU_PI,AR and CASIA face data sets.Compared with the direct use of all training sets for training,this method can reduce the cost of labelling by 30%-60%.Secondly,the cost-sensitive sequential active learning method is studied.The active learning is combined with the cost-sensitive sequential three-way decision.Each iterative process of active learning is used as a decision step of the face recognition system,and classifying each step according to the Bayesian risk minimum principle.The experimental results show that the proposed method can reduce the misclassification cost and improve the recognition accuracy of important samples,which is suitable for cost sensitive identification in real-world problems.Finally,considering the test cost loss in the decision process,the misclassification cost of the classification decision model is integrated with the test cost,and the total cost loss of the classification decision is obtained to be the objective function,then we can train a classifier with the smallest total cost loss.By combining active learning with cost-sensitive sequential three-way decision,the cost of labelling and misclassification of face recognition system can be reduced,and the active learning algorithm is terminated at the lowest total cost,which provides a new condition for setting the termination condition of active learning,which is of great significance for improving the practicality of the recognition system.