View Full Version : "Zoom and Enhance" to Become a Reality Thanks to Machine Learning

11-28-2016, 09:22 PM

The one phrase from television that makes IT people and creative professionals cringe the most is "zoom and enhance" - the notion that you zoom into a digital image and, at the push of a button, it converts a pixellated image into something with details - which lets CSI catch the bad guys. Up until now, this has been laughably impossible. // Google is tapping into machine-learning, in an attempt to change this.

11-29-2016, 03:16 AM
Just as long as it doesn't make all those old-fashioned clunky computer noises Mr Scott likes to add to his movies I'm all for it,so now its just down to getting my hands on a box fresh Nexus 6 Rachiell replicant I guess

11-29-2016, 10:56 AM
No information can be created. But there might be some recoverable information in the diffraction patterns and diffusion created by an invisible structure X, such as an edge. The algorithm has to make a guess and assume that the structure X is there, then create a model of diffraction/refraction/diffusion, actually some sort of local ray tracing assuming the structure X, and then if the end result resembles the original pictures, you can indeed include structure X in the 'improved' picture. If the end result assuming structure X does not resemble the original picture, then the computer has to learn for a better structure.

But the algorithm can only make educated and not too risky guess, so for example how on Earth a computer could guess the presence of a Replicant in the room from its reflection in the picture of a defocused mirror, in Blade Runner ? This trick didn't work on me when I saw the movie, but it was a good movie anyways. I wish they had gone further on the issue : 'Rick Deckard' cannot figure out if Rachel is a Replicant or Not, so how can he be sure he is not a Replicant himself ?
Perhaps in the next Blade Runner ?

11-29-2016, 12:03 PM
I would not consider a 'guesstimate' which is essentially what we are talking about. As evidence. It is no more solid than hearsay. If the source data only contains pixelated blurry noise, you cannot get any type of evidence from it. Just because that guesstimate happens to look like someone, is pure conjecture, and must be dismissed.

11-29-2016, 01:03 PM
This was teased as a photoshop tool in 2011, but it never saw the light of day. It's very suspicious. https://www.youtube.com/watch?v=XLLJBfRzXIQ&list=PL6D28B2A8CA3A2EB4