MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202621092295 A) filed by Symbiosis International Deemed University on July 29, 2026, for Forensic Facial Image Generation And Criminal Identification System Using Latent Diffusion Models.
Inventors include Prof. Riddhi Doshi; Arpita Waghmare; Sampada Bandre; and Lavanya Tiwade.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: ABSTRACT FORENSIC FACIAL IMAGE GENERATION AND CRIMINAL IDENTIFICATION SYSTEM USING LATENT DIFFUSION MODELS The present invention discloses a forensic facial image generation and criminal identification system (100) using latent diffusion models. The system comprises a text processing module (110) employing a CLIP text encoder (114) for converting natural language witness descriptions into embedding tensors, an image generation module (120) utilizing a latent diffusion model with a U-Net denoiser (124) operating in a compressed latent space for computationally efficient image synthesis, a dual-mode prompt engineering module (130) for simultaneously generating photorealistic mugshots and forensic pencil sketches with structured negative prompts, a face matching module (140) employing an ensemble of deep learning face recognition models (142, 144) with cosine similarity scoring for biometric comparison against a criminal database, and a database management module (150) with session persistence. The system achieves a 48-fold computational reduction over pixel-space diffusion, produces dual forensic outputs from a single text input, and enables automated suspect identification through calibrated similarity thresholds. [
Disclaimer: Curated by HT Syndication.
