Person identification using fusion of deep net facial features

Abstract

Face base identification is the method of recognising individuals through face images having application domains like information security and surveillance system. Deep networks have proved to be successful for facial identification in which entire images are passed as input to deep-net, and the network does feature extraction as well as classification. However, such a process requires millions of images to work with and implementing the same sometimes becomes complicated and time-consuming. This paper utilises pre-trained deep networks Alexnet and VGG-16 as feature extractors in which contribution to more significant level layers are utilised as feature vectors. The feature vectors are then diminished using a combination of PCA and LDA. After the reduction in the dimensionality of highlight vectors, they are intertwined and characterised using support vector machines. The proposed framework is assessed on VIDTIMIT database, highlighting the performance as far as precision, accuracy, and recall.