Tuesday, March 19, 2013

Better training leads to better face recognition

AUSTRALIA: Crims outed by Vic cops' facial recognition system
Key to the accuracy of the system was the composition of photos according to strict positioning criteria.

Victoria Police senior sergeant Cameron Tullberg said the quality of suspect photographs had degraded over decades.

At a recent police technology conference in Melbourne, Sgt Tullberg showed fellow officers recent photographs of such low quality that identification of suspects was almost impossible. In one photo an entire face was cropped out.

He said the requirements had been "turned all the way up", forcing police officers to properly compose a photo before it would be accepted by the system.
We have often made the point that facial recognition systems are best used by skilled operators. Their operation is far more complex than, say, fingerprint systems. This story from Australia draws attention to the fact that sensitivity to facial recognition "best practices" on the front end (data-gathering phase) leads to better matching down the road.