A Face Recognition System to Distinguish Monozygotic/Identical Twins

Face recognition systems often fail to distinguish monozygotic twins, commonly called as the identical twins, due to their high facial similarity. Specifically, changes in illumination or expression reduce the discriminative ability of such systems further. Hence, this research introduces a novel twin face recognition methodology that combines Super-Resolution Convolutional Neural Networks (SRCNN) and Multi-Orientation-Summed Gabor Local Binary Patterns (MSGLBP) to achieve twin recognition across four test cases, which involve real-time constraints such as illumination variations and changing facial expressions. At first, the SRCNN module of the proposed method enhances the high-frequency details and the texture quality in the face images of twins, so as to improve recognition under such constraints. Subsequently, the subtle biometric differences between the twin faces are extracted by capturing local texture variations across multiple Gabor orientations using MSGLBP. Finally, identity recognition is done using Support Vector Machine (SVM). Experimental outcomes confirm that a recognition accuracy and an F1 score of up to 98.48% and 0.98, respectively, are achievable with this method using the ND-TWINS-2009-2010 dataset. This research contributes to advancing face recognition technology and ensuring security in Human Computer Interaction (HCI), particularly in challenging scenarios involving identical twins.