It is my biased opinion that
http://www.imagemagick.org/discourse-server/viewtopic.php?f=1&t=21660 http://www.imagemagick.org/discourse-server/viewtopic.php?f=1&t=21695 which I intend to program into GEGL over the next few months, are generally better. (Note that neural net/fractal methods are better when images have self-similarity, that is, patterns repeated within the image, possibly over multiple scales, hence the choice of some of the test images at the Weinsman Institute site. Not all images have such self-similarity.)

David Thurgood spent a good amount of time comparing "my" recent schemes to the SR scheme of Daniel Glasner, Shai Bagon and Michal Irani (http://www.wisdom.weizmann.ac.il/~vision/SingleImageSR.html) and, in his unbiased eye, the SR scheme is the clear winner.

On 19/08/12 15:26, David Thurgood wrote:

Dear Sir,
I apologise, in advance, if this is the wrong e-mail address to use, but I confess to finding the GIMP lists rather confusing. Please forward this to the most appropriate GIMP developer.

The reason for writing: I have just seen this most interesting site that shows the results of a new processing algorithm for picture enhancement. The results are impressive indeed! So it occurs to me that your already excellent GIMP programme may one day have an extra plugin.

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

Some of those enhancements are incredible (in a literal sense of the word). specifically the final MC Escher(esq) towers.

If you look at the fences under the smaller arches, they have been interpolated from nothing. does the algorithm somehow spot the similar regions at other scales, and filled in the small arches with copies of the big arches? It has also been smart enough not to make sharp details out of the compression artefacts, for example around the top of the arch on the light side of that largest tower.

sam

On Thursday, August 23, 2012, 0:01:04, sam tygier wrote:

Some of those enhancements are incredible (in a literal sense of the word). specifically the final MC Escher(esq) towers.

I found the eye chart the most incredible:

Look at the bottom row.

sam tygier (samtygier@yahoo.co.uk) wrote:

If you look at the fences under the smaller arches, they have been interpolated from nothing. does the algorithm somehow spot the similar regions at other scales, and filled in the small arches with copies of the big arches?

Apparently so:

"Our approach is based on the observation that patches in a natural image tend to redundantly recur many times inside the image, both within the same scale, as well as across different scales. Recurrence of patches within the same image scale (at subpixel misalignments) gives rise to the classical super-resolution, whereas recurrence of patches across different scales of the same image gives rise to example-based super-resolution. Our approach attempts to recover at each pixel its best possible resolution increase based on its patch redundancy within and across scales."

Really cool.

Bye, Simon

http://en.wikipedia.org/wiki/Self-similarity

SR and Fattal are most effective when the image contains much of it, and this is reflected in the choice of test images.

On 22 August 2012 20:10, Nicolas Robidoux wrote:

http://en.wikipedia.org/wiki/Self-similarity

SR and Fattal are most effective when the image contains much of it, and this is reflected in the choice of test images.

Which also happen in most images in the real world, as denoted by Dr. Mandelbrot.

But'd guess the algorithm also takes advantage of other pixels on the same edge (i.e. the self-similarity across pixels on an object boundary, not necessarily on
a different scale)

Either way - it _is_ cool! Are the details on the paper enough for us to get an implementation of this started?

Joao:

Read the article!

My guess is that the SR scheme combines two things:

1) Look for "similar things" at the same scale within the image to reconstruct a "platonic ideal" which you then replicate (this is super-resolution using data from the same image; classical super-resolution uses multiple shots of the same scene). 2) Do the same at multiple scales, in the hope that there is self-similarity to pick up on.

But I've only read the summary.

-----

Prequel: I am very biased because I develop and program competing methods which could be described as belonging to "older" generation approaches. So, I'm fighting obsolescence.

Now: If I was to suggest a "state-of-the-art cool" method to someone, I would suggest NNEDI3 (which is programmed in Avisynth).

Why?

a) NNEDI3 not an academic scheme. It's well tested in real world situations, I'm pretty sure that the code is FLOSS, and there is an open source community around it. It was put together so that it does pretty well all the time, not only when the data fits the "ideal input".

b) No matter how wonderful the "discovery" of how common self-similarity is in this world (a crush with fractals actually contributed to getting me back into grad school a few years back), the world is not universally self-similar.

Take out the test images from the Weinzmann site in which you do not find repetitions of the same pattern, either at the same scale or at multiple scales. Does the SR method really create a faithful or better enlargement?

(Bias warning.)

If you look for self-similarity, you'll find it...where it's not.

Look at the skin of the baby. Do you see that the woolen hat was "woven" into it?

And I really dislike what SR does to eye pupils.

Also: Some of the sharpness comes from applying something which in the end is a lot like a variation diminishing limiter (Jensen like?). Do you really want this "waxy look"?

Not that these things could not be fixed (it's most likely a matter of setting thresholds).

This being said: Do my methods do better? Maybe not. But they are local (SR requires an analysis of the whole image), fairly cheap, and adapt to large enlargements or reductions robustly.

-----

Again, in the same ballpark, it is my opinion that NNEDI3 (which I have absolutely no connection with) is likely to disappoint less.

Before starting to program SR, I would also contact the authors and ask

1) If they are, or know people who are, willing to help

2) If there are patents that get in the way.

On 23 August 2012 08:55, Nicolas Robidoux wrote:

Before starting to program SR, I would also contact the authors and ask

1) If they are, or know people who are, willing to help

2) If there are patents that get in the way.

On the other hand - how is your work going? Are you making it avalilable as a GEGL OP as well?

regards,

js ->

On the other hand - how is your work going? Are you making it avalilable as a GEGL OP as well?

Yes, that's always been the plan. But everything else I do has been taking a lot longer than expected...

At least, I think that I've come to the point where my methods are just about as good as they can be. Exploration is winding down.

One last grumpy comment: Look at the right nostril of the baby. It's pure fiction.

Sorry: Left nostril.

On 23 August 2012 09:07, Nicolas Robidoux wrote:

One last grumpy comment: Look at the right nostril of the baby. It's pure fiction.

Since we are to this - one of the thigns that had impressed me the most is the border of the yellow helmets of the workers - on one of the last pics. I certainly would like to have the ability to get to such a result within an app on my desktop, even if it depended on parameter tunning.

But - again -- I am looking forward for what you are doing there! :-)

(as for the "check if there are software patents in the way" -- unfortunately it is not 1995 anymore, where one or other "lzw" patent could be avoided.

It is 2012 - and there are software patents even to blinking your eyes and clean your b. -- it is just a matter of whether the patent owners target you or not.

We are in a point where either all software patents will collapse into a legal singularity,
or programming anything will be illegal. )

js
->

_______________________________________________ gimp-developer-list mailing list
gimp-developer-list@gnome.org
https://mail.gnome.org/mailman/listinfo/gimp-developer-list