Face recognition using convolutional neural network with proper feature extraction and combination
Paper ID : 1150-MVIP2020
Javad Khosravian Arab, Kourosh Kiani *
The last 25 years have witnessed a great advance in the field of face recognition Despite this fact face recognition is still a challenge. This paper proposes a deep learning-based approach that exploits a convolutional neural network.In this approach, the first step starts with feature extraction using ResNet101.These features considered as an input of a block in pyramidal structure for combing the feature and extracting new features which brings a general view And the use of convolutional and deconvolutional networks to combine higher-level features of the image to extract more general features. Our approach is trained in two large-scale datasets using three loss functions and tested on the LFW dataset to have a comparison with previous methods. The results show a precision of 99.77 % accuracy in the LFW dataset.
Face recognition, deep learning, pyramid structure, convolution and deconvolution
Status : Paper Accepted (Oral Presentation)