Depth Face Recognition through Deep Learning Networks Fine-Tuning
Face recognition still has many challenges, and with the appearance of the Microsoft Kinect device, new
potentials of research were discovered, trying to use the Kinect depth maps as a data source to recognize human faces was
considered an interesting research area, mainly because the Kinect could provide the required data in an affordable and
accurate way, but till this day no research managed to utilize the new and popular deep learning techniques to achieve higher
accuracy and better results on face recognition using any deep learning technique and Kinect 2 depth map images. In this
paper, with the goal of enhancing face recognition through the use of depth map images and the lack of a depth map datasets
that can be used to train a deep learning network, we introduce a new practice for network fine-tuning, as the methodologies
were applied with face depth maps to achieve high face recognition accuracy. The process of building a convolutional neural
network and loading weights of a similar Image-net trained network was introduced, where the network was fine-tuned and
trained to work with depth map images of faces. The results of the new training procedure presented superior performance
compared to any previous methods where depth maps were the source of data.
Keywords- Face recognition; Kinect 2; Deep Learning; Fine tuning, Convolutional neural network.