Age Invariant Face Recognition using Deep Sub-Pixel Resolution Features

Age-invariant face recognition has many real-world applications. Despite significant advances in this field, it is still challenging to accurately predict faces across various ages as a person’s age changes the face significantly over time, leading to a lot of intra-class variations. In this paper, we have attempted to use sub-pixel interpolation over an extracted face, as a pre-processing step for increasing image resolution, for age invariant face recognition. We have employed the Xception deep learning architecture over the Casia - Webface dataset for our experiments. We show that there is a significant enhancement in recognition accuracy when sub-pixel interpolation is used compared to when images are given to the deep learning model without any pre-processing.