Third Eye: AI-Based Forensic Sketch Drawing and Face Recognition System

When a crime leaves no photograph or video of the offender, investigators often have nothing but a witness’s memory to work with. Turning that memory into a face normally takes a trained sketch artist and a good deal of time. This paper describes Third Eye, an AI-assisted system that brings together steps which are usually done separately: taking a witness description as text or voice, drawing a forensic sketch, searching a database of faces, and documenting the case. Speech is first converted to text, and natural language processing then pulls out facial attributes such as face shape, eyes, eyebrows, nose, lips, hair, beard, and moustache. The system suggests a facial component for each attribute, and an investigator reviews these suggestions before the sketch is put together. The finished sketch is cleaned up, turned into a face embedding with a model such as ArcFace (or an equivalent), and compared by cosine similarity against an authorized demonstration database, which yields a ranked list of candidates. Case records and a PDF report are saved for later review. Third Eye is meant to support investigators, not to decide identity on its own. This draft does not yet include datasets, implementation settings, or measured results; these are marked [To be filled] rather than estimated. In today's world, identifying suspects in criminal investigations is a challenging task, particularly when no photographs, video recordings, or direct visual evidence of the offender are available. In such situations, investigators often depend on eyewitness descriptions to reconstruct the appearance of a suspected individual. Traditional forensic sketching methods rely heavily on the experience and artistic skills of forensic sketch artists, making the process time-consuming and sometimes difficult when witness descriptions are incomplete, unclear, or inaccurate. To address these challenges, this paper proposes Third Eye: An AI-Based Forensic Sketch Drawing and Face Recognition System, an intelligent framework designed to assist investigators in generating facial sketches and identifying potential suspects using Artificial Intelligence, Machine Learning, Natural Language Processing, and Computer Vision techniques. The proposed system provides an integrated approach that converts textual or voice-based witness descriptions into structured facial information. Speech recognition techniques are used to convert spoken descriptions into text, while Natural Language Processing (NLP) extracts important facial characteristics such as face shape, eyes, eyebrows, nose, lips, hairstyle, beard, and moustache. These extracted attributes are processed to identify suitable facial components from the available component library.