Reconstructing a Face Through Time

An intriguing question is whether a computer can learn not only what a face looks like today, but how that face may change over many years. Human ageing is not random. The proportions of the face change during childhood, while later life brings more subtle changes in skin, soft tissue and facial structure. If these patterns can be learned from large collections of photographs taken at different ages, it may become possible to generate a plausible future appearance from an earlier image.

The most important application could be in missing person investigations, particularly when someone has been missing for many years and the available photograph no longer resembles their present appearance. A computer assisted ageing system could produce updated facial images to support public appeals, police investigations and comparisons with newly discovered photographs. Rather than relying entirely on an artist's interpretation, statistical models and machine learning could learn recurring ageing patterns while preserving those facial characteristics that are most closely associated with identity.

There are wider investigative possibilities as well. Age progression and regression could help compare photographs believed to show the same individual at very different stages of life, or assist facial recognition systems when the time gap between images is large. The challenge is that there is no single way in which a person ages, so such images must be treated as probabilistic reconstructions rather than predictions. The real opportunity is to use computation to narrow the search and give investigators new visual evidence where previously there may have been little more than an old photograph.