Machine Learning Based Approach for Person Identification in Group Photos

Abstract

In this digital era, capturing photos using smartphone camera is very handy. Especially when we are hanging out with friends or family or attending a wedding and so on, we end up taking many group photos. But when browsing through these group photos, most of the times, a person is interested in only those photos in which he himself is present. So, currently we manually browse through such group photos in the phone gallery and then identify the pictures in which the specific person is present. For group photos, this procedure needs to be repeated for every person in the group. In this paper, we have designed and implemented an Android platform based photo grouping application named “EuphoriaGrouping” (EUG) using Neural Networks. EUG application automates the process of detecting faces and identifying persons from a group photo. It maintains a catalogue of group photos for every person present in the group photo. For person identification, two different convolutional neural network models, viz, Custom Built and OpenFace CNN are used. Implementation and performance comparison of these models is presented.