A way to improve precision of face recognition in SIPP without retrain of the deep neural network model
share
Although face recognition has been improved much as the development of Deep Neural Networks, SIPP(Single Image Per Person) problem in face recognition has not been better solved. In this paper, multiple methods will be introduced to improve the precision of SIPP face recognition without retrain of the DNN model. First, a modified SVD based method will be introduced to get more face images of one person in order to get more intra-class variations. Second, some more tricks will be introduced to help get the most similar person ID in a complex dataset, and some theoretical explain included to prove of why our tricks effective. Third, we would like to emphasize, no need to retrain of the DNN model and this would be easy to be extended in many applications without much efforts. We do some practical testing in competition of Msceleb challenge-2 2017 which was hold by Microsoft Research and finally we rank top-10. Great improvement of coverage from 13.39% to 19.25%, 29.94%, 42.11%, 47.52% at precision P99(99%) would be shown in this paper.
READ FULL TEXT