Modern mobiles like the P20 Pro have 8/20/40 MP cams and mix them AI-based. For picture forensics a multi-layer naming is required.
RR Real Reality (physical 2D layer catches 3D world)
AR Augmented Reality (RR plus Google)
VR Virtual Reality (no RR digital generated all)
MR Mixed Reality (RR plus Microsoft Cortana)
Physical cams of mobiles (skip of selfie-cam)
SC Single Cam
DC Double Cam
TC Tripple Cam
There is an increasing distortion between RR and TC which is destroying picture forensics and risks to bring the false people behind bars.
How can LEO rebalance or compensate this trend?
Rely on facts from trusted cameras ?!
Reverse Image to RR engineering for crime evidence we want to understand. Its not picture forensics with user interaction.
All without any! user interaction.
Out of picture forensics we have over the years a bunch of examples which are visually not clearable by a low probability. They remain uncertain. It would be nice to get out of thousands of mobile cams a distortion factor by brand and by model to take into consideration for unclear pics.
The first triage is always 'photoshopped' or 'not' (our slang). But having pictures out of an eUFM and considering the cam distortion factor detailled split in 5 parameters we could better run Reverse Image Engineering.
Something like an USGLDCM (Universal SigGraph-like Distortion Correction Model), i.e. this approach extended beyond portraits and selfies? ?
http//
applying to images standard tensor-vector multiplications in the in the i-th dimension is definitely an interesting field of research.
Besides the paper, the good guys at Princeton have also a demo site
http//faces.cs.princeton.edu/
jaclaz
No, its more complex as AI is the 'new kid in town' (we started with the iPhone X last year).
No, its more complex as AI is the 'new kid in town' (we started with the iPhone X last year).
Well no, allow me to disagree.
AI is not a new kid (kids actually exist and - within their limits - tend to work pretty well), AI is just one of the latest empty (or containing everything, you choose) buzzword.
There is nothing actually artificial, let alone intelligent in 99% of things that are advertised as AI, only good ol' programming by humans.
jaclaz
How can LEO rebalance or compensate this trend?