Reference date: 30 October 2026 — research paper published; 13 November — university summary released

Keele University highlighted a facial-image detection study on 13 November that combines ordinary colour information with measurements of brightness patterns. The FDIP-NET model draws mathematical inspiration from fluid dynamics without simulating fluids. The underlying research paper was published on 30 October.

Tests averaged across three training runs produced accuracy between 98.654% and 99.669% on three facial-image datasets. Each model was trained and tested within its own dataset. An additional evaluation on individual FaceForensics++ video frames achieved 91% mean accuracy.

The authors stress that strong results within known datasets do not demonstrate reliable transfer to unfamiliar image collections. The research also examines prediction confidence and associations between errors and properties such as blur, brightness and contrast, identifying questions for further work.

Sources

  1. Keele University: 13 November research summary ↗
  2. Original research: FDIP-NET methods, benchmarks and limitations ↗

Original summary based on the sources above, checked for the November 15, 2026 edition. Reference dates are shown separately from the edition date.