EFIT-V and personalised avatars
EFit-V is one of the newest tools in the police arsenal. A replacement for sketch artists and the electronic EFit system. Rather than concentrate on how wide a nose is, or how far apart the eyes are, whilst it still has that capability, the software takes a new approach, and creates photorealistic faces whilst it is at it.
That EFit-V has potential for avatar creation should already be obvious.
The software uses feature-based composites, similar to E-fit - it can select eyes, brows, noses, mouths et cetera, from a database, and paste them onto a face. However, its real strength is that it can go so much further than that. It uses genetic algorithms to make a selection of several different faces based on the details that can be remembered. When used to help solve a crime, the victim looks at the face which most closely fits the perpetrator, and selects it. This face is then used to seed the next selection of nine new faces, created by the program varying the genetics of the original..
Once all the features are in that can de determined - the whole face is not necessary - the program matches and blends skin tones, and builds a photorealistic 3D face from them, and each of the slightly different offerings are also just as photorealistic. Should an issue such as the resulting skin colour turn out to be wrong, that can be adjusted on the fly, always zeroing in on the ideal face; that which is in memory. By using this approach, it takes the emphasis away from individual features, and more towards looking at the whole face, in as much detail as possible.
The software can also start from scratch, working without facial feature descriptions, and generating its own faces that progressively evolve to match the witness' memories. The witness starts with a general description such as "I remember a young white male with dark hair." Nine different computer-generated faces that roughly fit the description are generated, and the witness identifies the best and worst matches. The software uses the best fit as a template to automatically generate nine new faces with slightly tweaked features, based on what it learned from the rejected faces.
The mathematics underlying the software is borrowed from Solomon's experience using optics to image turbulence in the atmosphere in the 1990s. "I then realised that the same technique could be applied to human faces, which in many respects are mathematically similar to turbulent wave fronts," said Solomon.
Of course, the software has many uses beyond face recognition. Given that it creates photorealistic faces, and can be compared to photofit ids - which by their nature are full frontal and side on - the software is capable of creating tweaked and completely unique faces that match user desire, but at the same time seem completely real. Given the increasing number of VR environments that are using FaceGen or a similar approach, to create realistic avatar heads from photographs, the facility EFit-V offers cannot be ignored.
The software has now started to make its way to the United States, where it being used by researchers in university settings.
EFIT 5 Flyer (PDF)