Algorithms to enhance or transform images | Hacker News Reader

Algorithms to enhance or transform images

scriptol.com

235 pointsscriptproof38 comments

Some of these (esp beautify) would make great 99c iphone apps

I was about to call BS on a few of these, but then clicked through to the detailed explanations and I'm pretty damn impressed. Colorization technique and UI is pretty slick.

Agree - the colorization did it for me as well. Very nice.

I remember reading the original paper years ago. http://www.recolored.com/ is based off that research.

I'm surprised content aware image resizing didn't make the list. It is a few years old but still pretty neat:

http://www.youtube.com/watch?v=qadw0BRKeMk

That deserves an update, yes.

The "remove an object" image transformation reminds me of the Photoshop new feature "content-aware fill" featured here: http://youtu.be/NH0aEp1oDOI?t=3m33s

If I recall correctly, they both use the same inpainting algorithm.

That's right AFAIK, it's PatchMatch http://gfx.cs.princeton.edu/pubs/Barnes_2009_PAR/index.php

A discussion on the pixel art scaling algorithms was going on here, if you're interested in them: http://news.ycombinator.com/item?id=2034690

Any actual implementations of the pixel art scaling algorithm available?

http://scale2x.sourceforge.net/download.html

edited to remove snarky comment

scale2x ≠ the one featured in the submitted URL’s roundup.

Strange, he was asking about the wiki article.

Even though I've interned in the field of image processing (and actually understood the math after a while!), the emerging results never cease to amaze me.

The 'beautify a face' transformation should be renamed 'normalize a face' or something slightly less subjective then beauty. Although, I suppose 'normalize' isn't any less subjective. Suggestions? Perhaps 'standardize' or 'anglicize' or 'fashionize'?

Human opinion of what's "beautiful" in other human faces is not entirely subjective. It has been know for a while that if you start averaging faces you end up with a "pretty" face. This is because we genetically like symmetry, and averaging out faces makes them very symmetrical.

There was a paper here on HN a few months back that compared averaged faces to faces with unchanged bone structure but smoothed skin; the differences in rated attractiveness were apparently negligible (and average faces with bad skin faired poorly). Averaging removes blemishes.

Cartoonists and animators have long contended that some small asymmetries enhance the emotional connection we have with a character --- think of a puppy with one ear that points up, or with a spot over one eye. Sympathy isn't beauty, I admit; I think it might contribute to attraction, though.

some small asymmetries enhance the emotional connection

Small asymmetries. Which, by design must be built on top of something that is otherwise relatively perfect. Otherwise the asymmetry does not stand out, and the whole thing is just a mess.

For example, a mangy old cur with one ear that points up, or a spot over one eye.

Perhaps, but the use of a term like "we genetically like symmetry" bugs me. Has it actually been scientifically proven that humans have a genetic predisposition toward preferring symmetry? My google-fu turned up nothing. Isn't opinion naturally subjective, albeit informed by societal influences one way or the other?

My google-fu also fails me. This is the best I found: http://www.metafilter.com/62173/The-hypnotic-beauty-of-the-a...

I based my initial opinion on a documentary about someone who had tried to find the "typical criminal" face by averaging the mugshots of criminals. But to his shock, the resulting face kept turning out beautiful. But again, my google-fu fails me.

This isn't symmetry per se, but the study referred to here does suggest that beauty is mostly innate:

http://www.newscientist.com/article/dn6355-babies-prefer-to-...

(They didn't do any averaging.)

This article says that symmetry specifically has also be found to be attractive to babies:

http://www.jyi.org/volumes/volume6/issue6/features/feng.html

Not sure if that was separate from the averaging mentioned.

>Has it actually been scientifically proven that humans have a genetic predisposition toward preferring symmetry?

http://www.grajfoner.com/Clanki/Perrett%201999%20Symetry%20A... seems to imply so. (I'm sure you can find something else stating the opposite though.) That paper was cited as a reference in the original beautification paper, which can be found here http://www.leyvand.com/research/beautification2008/attractiv....

The explanation I heard once was that childhood illnesses have a large impact on symmetry. In general, symmetry implies health/fertility, so it makes sense that we're attracted to it.

"beauty" relates to desire and reproduction. Simmetry means healthy genes. A gene that makes you see "beauty" on external signs of good genes is a gene that gets naturally selected. So I guess most of the actual population have it and see beauty on simmetry.

I remember an art teacher who demonstrated me that we actually don't like symmetry in faces - he shown us a picture of a woman face composed of one half, and its mirror image. This face was completely symmetric (vertically), but didn't feel quite "right".

That seems a good deal less scientific than the numerous studies that have strongly suggested the opposite.

You are right, of course. I was just providing a data point.

But the face on the left is more beautiful. She has large eyes and lips and a strong jaw -- outstanding features which arrest the gaze. The "beautified" version of her face was just kind of "meh".

I think most people would disagree with you on that.

> [from the description on the site] There is no written rule to define whether a face is beautiful or if it is not, then the creators followed the principle that what is more closer to the common is better.

I initially thought this was the same experiment done a while ago where people rated different faces according to their preferred beauty. Then the algorithm would normalize an input face based on those preferences.

Normalizing just to the common face and having it become more beautiful is kind of unintuitive to me. For instance, I wonder what it would look like to run that algorithm on an already generally-accepted beautiful face and see if it became less so.

I honestly think you could make that argument for the pictures shown. But that's just my subjective opinion...

I see now that the details state that "It is possible through an algorithm for a computer to process a photo or video and produce a modified image, possibly improved."

So the use of "possibly improved" makes me happy. I suppose the naming of the algorithm with 'beautify' helps to describe what it typically would be used for.

A software service with an API that performs these transformations would be amazing.

...or, a disorganized regurgitation of papers that have been posted to HN several times in the past.

I know it's tiresome to complain about reposts, but can we at least not link to cut-and-paste content farms?

The face beautification looks like two entirely different faces.

Some of these are quite cool, particularly the beautification algorithm. But where can these algorithms be found...?

this appears to be blogspam.

Reply on news.ycombinator.com