Deep Learning Approach for Human Face Feature Analysis

Abstract

In law enforcement, hand-drawn facial sketches from eyewitnesses are often used when photos are unavailable, but matching them with real images is difficult due to visual differences. Third Eye addresses this by providing an intelligent sketch- based face recognition system. Investigators can create composite sketches using a drag-and-drop interface with modular facial features. The sketches are processed using CNN s and matched with photographs via embedding comparison methods like Siamese networks. AWS Recognition handles cloud-based face matching, and SQLite manages local data. The system combines Python (Open CV, PyTorch/TensorFlow) for deep learning and Java- FX for the front-end, enabling fast and accurate suspect identification.