Topic awaiting preservation: Super Resolution

Got this from Doug over at Retouch Pro.

http://www.extremetech.com/extreme/132950-csi-style-super-resolution-image-enlargment-yeeaaaah

and

http://www.wisdom.weizmann.ac.il/~vision/SingleImageSR.html

Ooo lovely stuff, thanks war J shall have to investigate.

I have been doing some more reading here-n-there, but nothing too in-depth. You know, just scan and get the basic idea(s). The above examples seem to be based on patches found within the same photograph. In the paper, PDF at top of second link, there is an example with shrubery with the patches shown.

One method in particular that caught my eye is based on gradients. Might look some more into this.

And then there are Bayesian methods, which are beyond me. Bell curve is about all I know about probability and statistics. Rock on, Gauss.

Now, the one that really caught my eye, is a learning-based methodology. Train and watch the magic. I have a thing for neural nets and genetic algorithms, so might be right up my alley. But I won't know for sure until I do some more digging.

Looks amazing. Can't imagine this getting much better anymore. The resulting images look very similar to what you'd expect from a medium level digital camera 100% crops under normal conditions. Which means they're usable for 90% of possible purposes.