Recognition of identical twins based on the most distinctive region of the face: Human criteria and machine processing approaches

Abstract

Face recognition domain has been well advanced and has achieved high accuracies in identification of individuals in recent years. But in practice, distinguishing similar faces such as an identical twin still is a great challenge for face recognition systems. It happens due to very small differences in the facial features of them. Therefore, extracting common face features is not proper for differentiating identical twins. A solution to this problem is to find the most distinctive regions in the face of identical twins. In this paper, two procedures used to find these specific regions: 1) Machine Processing: A Modified SIFT (M-SIFT) algorithm has been implemented on Identical twins’ face images. Each face image has been segmented into five regions contain eyes, eyebrows, nose, mouth, and face curve. The location and number of mismatched keypoints represented the most distinctive face region in the face of identical twins. 2) Crowdsourcing: We have recognized differences between identical twins faces from human criteria viewpoint by enlisting crowd intelligence. Several questionnaires were designed and completed by 120 participants. The dataset of this study collected by ourselves and include 650 images for 115 pairs of identical twins and 120 non-twin individuals. The results of Machine Processing and Crowdsourcing methods showed that the face curve is the most discriminant region among every five regions in most of identical twins. Several features proposed and extracted based on the keypoints of the M-SIFT algorithm and face landmarks. The experimental results demonstrated the lowest equal error rate of identical twins recognition as 7.8, 8.1 and 10.1% for using the whole images, only frontal images and only images with PAN motions, respectively.