FACIAL VERIFICATION OF FAMILIAL RELATIONSHIPS USING DEEP LEARNING AND SIAMESE NEURAL NETWORKS

Abstract

Facial kinship verification is all about figuring out if two people are related by blood by looking at their faces. This task is different from traditional face recognition because it looks for small similarities that may exist between different people in a family. This work develops a deep learning-based method that uses a Siamese neural network to compare pairs of facial images and find out how similar they are in a learned feature space. The system processes input images through the same network branches and uses a distance metric to figure out how related they are, as well as a classification system based on a threshold. An uncertainty region is added to handle cases that are on the edge of being true or false. This makes the system more reliable.The system also uses explainable AI methods to give visual and numerical information about the prediction, such as heatmaps and region-wise analysis of facial features like the eyes, nose, and mouth. The model is tested on datasets like KinFaceW and Families in the Wild. It works well on structured data, but it shows problems when used in the real world. The findings suggest that integrating similarity learning with interpretability can yield a more transparent and effective approach for kinship verification.