Treffer: Sibling Face Recognition Improvised by Merging Deep Learning Embeddings from VGGFace and FaceNet

Face recognition between siblings is difficult because there is high class similarity and limited variability between faces. Though deep learning algorithms such as FaceNet and VGGFace have good overall recognition, they are not sufficient for sibling identification, particularly between certain regions on the face.To meet this, our research proposes a hybrid method of fusing FaceNet and VGGFace feature embeddings. Through their complementary strengths, we build a composite feature vector per image and calculate cosine similarity for classification.This fusion approach significantly enhances sibling classification accuracy over individual models. Experiments on HQf subset of SiblingsDB demonstrate increased precision and reduced misclassification in full-frontal and cropped facial regions. The work also points out how fusion eliminates regional bias, resulting in more equitable performance.Our fusion approach can be used in biometric authentication, forensic family searches, and kinship-based identity systems—areas where accuracy and reliability are crucial.Executed in Python on actual sibling datasets, empirical findings verify greater accuracy and uniformity through this blended model paradigm