Employee Face Recognition Scheme Using A Common Space Mapping Approach

In this work, we present a FaceNet based ‘two branch’ model for employee face recognition in low resolution images captured using substandard camera sensors. Our model involves a common space mapping approach using two deep convolutional neural networks (DCNNs) that map the low resolution and high resolution face images to a common space. The model is trained such that the distance between the two mapped images in the common space is minimized. Then, a logistic regression classifier is used to classify the mapped image by the identity of the employee. We show through simulations that the presented model achieves a recognition accuracy of 99.84%, 98.88%, and 95.53% on $36\times 36$, $24\times 24$, and $16\times 16$ resolution images, respectively, for 209 subjects. Furthermore, the proposed model has less space (90 Megabytes) and computation requirements making it suitable for systems having low computing power and memory.