Abstract

Face recognition domain has been well advanced in recent years and has achieved good accuracies in identification of individuals. But in practice, distinguishing similar faces such as an identical twin, still is a great challenge for face recognition systems, due to very small variations between face features of them. So using common face features is not proper for this purpose. One solution is to find regions that is more different between twins' faces. Our approach to find these regions is based on examination of two methods: First, segmenting identical twins face images into five regions contain eyes, eyebrows, nose, mouth and face curve. Then, implementation of a modified SIFT algorithm on face images, distinctive face features of identical twins in machine processing manner was investigated. Second, we have also recognized differences between identical twins faces from human criteria viewpoint by using crowdsourcing paradigm and enlisting crowd intelligence. The proposed methods were tested on 110 pairs (220 individuals) of identical twins' face images. The results showed that face curve is the most discriminant region among every five regions from machine processing viewpoint and based on human criteria, in 65% and 35.5% of identical twins, respectively. As a result of this study, we have shown that face curve is the most distinctive region in identical twins' faces. Therefore, giving more value to the features extracted from the face curve region can be effective in improving the performance of face recognition systems.