Twins and Similar Faces Recognition Using Geometric and Photometric Features with Transfer Learning

Abstract

Twins recognition and identification is one of the important challenges in the field of image processing. The strong similarity between identical twins makes it hard to distinguish the twin from his/her sibling. Similarities come from biometric, geometric, and photometric features. In biometric patterns, the fingerprints found to be identical in some cases, geometrically, the twins’ faces rarely differ which confuses people. Photometric features are very close to each other even though they rarely success in twins’ recognition. We tackle this challenge by a model for twin’s face recognition (FR) where our solution is based on deep transfer learning in terms of residual neural networks including two VGG16 trained networks, which are considered to be one of the powerful and deeply learned neural networks. For comparison purposes, we check other approaches to solve the twins’ problem including iris, fingerprints, and lip corners. The data used was collected from Google which is a challenge. Data contains 4-pairs of twins with the 17-different position for each one which produces 5×2×17 (170) different images. Collected images were used for comparisons between features. Results show that geometrical features gave 85% of success while photometric features gave 96%. By hybridizing geometrical and photometric features together, the results reach 98% of accuracy. Biometric measures, in this research, prove the superiority of deeply transferring learning over traditional methods. The newly achieved method could be replaced to assist authentication systems that fully depend on biometric features.