Face Recognition Systems (FRS) are increasingly targeted by morphing attacks, where facial features of multiple individuals are blended into a synthetic image to deceive biometric verification. This paper proposes an enhanced Siamese Neural Network (SNN)-based system for robust morph detection. The methodology involves four stages. First, a dataset of real and morphed images is generated using StyleGAN, producing high-quality facial images. Second, facial regions are extracted using Faster Region-based Convolutional Neural Networks (R-CNN) to isolate relevant features and eliminate background noise. Third, a Local Binary Pattern-Convolutional Neural Network (LBP-CNN) is used to build a baseline FRS and assess its susceptibility to deception by morphed images. Finally, morph detection and classification are conducted using the proposed SNN framework, which incorporates a novel feature fusion strategy based on Canonical Correlation Analysis (CCA) to enhance discriminative power. The model is trained and evaluated using publicly available Face Recognition Technology (FERET) and Face Recognition Grand Challenge (FRGC) datasets, comprising 1,030 real and 2,000 morphed images. Experimental results demonstrate that the proposed method significantly strengthens the resilience of FRS to morphing attacks, achieving a high detection accuracy of 99.9%. This confirms the model’s effectiveness in distinguishing between real and manipulated images with minimal errors.
