Ai-based Forensic Face Sketch Generation And Suspect Identification From Eyewitness Verbal Description

• Author(s):

  Manikanda Prabhu V | Abilash S | Premkumar S | Arun Kumar C

• Keywords:

  Forensic Face Sketch Generation, Speech-to-Text, Generative Adversarial Network, Attention-cGAN, Vision Transformer, Face Recognition, Cosine Similarity, Deep Learning, Criminal Identification, Multilingual Processing.

• Abstract:

  Suspect Identification Based On Eyewitness Descriptions Remains A Critical Challenge In Criminal Investigations Due To Memory Limitations, Language Barriers, And The Lack Of Reliable Visual Evidence. Traditional Forensic Sketching Methods Rely On Skilled Artists, Making The Process Time-consuming, Subjective, And Often Inconsistent. To Address These Challenges, This Paper Presents An AI-Based Forensic Face Sketch Generation And Suspect Identification System For Automated And Efficient Investigation Support. The Proposed Framework Integrates A Multilingual Speech-to-text Module Using Whisper To Convert Eyewitness Descriptions Into Text. These Features Are Processed Using An Attention-Based Conditional Generative Adversarial Network (Attention-cGAN) To Generate Realistic Forensic Face Sketches, Which Are Further Enhanced Into Photo-like Images. The Generated Faces Are Encoded Using A Vision Transformer (ViT) To Extract Feature Embeddings And Matched With Criminal Database Records Using Cosine Similarity. The System Supports Multilingual Input And Provides Automated Result Visualization With Similarity Scores. Experimental Results Demonstrate Effective Performance In Generating Identity-consistent Facial Representations And Improving Matching Accuracy. The Integration Of Speech Processing, Generative Modeling, And Feature-based Matching Makes The System Suitable For Real-world Forensic Applications.

---

Other Details

• Paper id:

  IJSARTV12I4104998

• Published in:

  Volume: 12 Issue: 4 April 2026

• Publication Date:

  2026-04-14

Download Article