[Award Winner] Bitmapped Images - The technical side of things explained.

[RobA]

And for your viewing pleasure, the gimp resynth plugin going the full 800% in one shot Not a good resize, but an interesting 70's acid trip none the less:


And here is the best I could do with some playing.

This was done with Image Analyzer fractal resize x 2 then a wiener resize x2, followed by a fractal resize x2. Pulled into gimp and added film grain.


-Rob A>

[mathuwm]

Someone is having too much fun

[Redrobes]

Thanks Mathuwm and Rob for those. I like that Genuine Fractals one. I would agree that it is the best of them all. Its artificially sharpened up those edges just where it was required. That really is some clever software there.

[Redrobes]

I have edited in the space I left in post #13 of this thread about B&W resampling.

[Redrobes]

In this thread I was talking about antialiasing and how occasionally you see the spokes of the wheels of a western movie wagon go backwards or sometimes even look stationary because the camera frame rate is similar to the action. Well I cam across this today - you will have to forgive the title of this movie but I have never seen a better example of temporal antialiasing - ever. Its excellent !

https://www.youtube.com/watch?v=jWIcVP6GRfw

[RobA]

Time to revisit this one...

A few more tools. Here is a demo of a technique called "smartedge". It only resizes by x2. I ran it three times to get the reference image up to the 800% we are comparing:

SmartEdge demo:


Also, I tried a demo of EnlargerPro by Bearded Frog. Here is the (watermaked in the demo) result:


As a final note, Apply sharpening after an image to the final size, not the other way around. Otherwise unperceivable sharpening halos may become clearly visible!

[Vandy]

Hello, Redrobes.

Thanks for a great tutorial. There is a tremendous amount of excellent information here.


FOR ALL INTERESTED

I've attached a PDF document I created from this tutorial. The only thing I've done is to reformat the tutorial and correct any spelling and grammical errors. The content has not been altered.



Enjoy.

Regards,

Gary

[Redrobes]

Well done, that's very nice. Spread it about a bit to anyone who might benefit from it.

The only issue is that some of the images need to be viewed without any stretching applied because they are describing what goes on when you stretch images. The PDF viewer is scaling them itself so just be mindful that its happening and will affect the images. The one in particular is the 997 to 1000 pixel stretch using all three types (3rd pic of post #11) where I argue (not very firmly tho...) that pixel resample is possibly the best in that particular case and in the reader they all munge into something which looks the same since there's another stretch being applied on top. Same goes for some of the dithering images etc.

I think in general its easier to read in your PDF than the original tho as the images are inline and bigger.

[thechao]

Three topics:
(1) Genuine Fractal sampling
(2) Gamma correction in saturated images
(3) Upsampling rasterized vector images

(1) Genuine Fractal Sampling
I was interested in the "Fractal" upsampling technique since the literature on image resampling doesn't contain this term --- the closest I could find were statistical techniques for preserving edges and Haar/other wavelet transformations. I found the website for the "Genuine Fractal" (GF) method, and it is interesting to see how it works. I believe, overall, the GF upsampling method is roughly the same as repeated bicubic upsamplings at factors of 2, followed by smoothing, and edge-sharpening to keep the image from blurring. This, effectively, automates the "by-hand" upsampling technique you describe earlier in this tutorial.

I have no clue why they call this "Fractal" sampling, since it does not look like any of the mathematics of fractals. Also, the method is patented, although, the patent is from the early/mid-90s so it is probably close to being expired.

(2) Gamma Correction
I'm not sure the exact colorspace, nor gamma correction method being referred to. However, usually gamma correction occurs in an unclamped colorspace, and samples to a 32-bit floating-point channel. By adding a gamma of .5 for a saturated image, then upsampling, then inverting the gamma to 2, you've applied a isomorphism to all the pixels that maintain their saturation levels (up to round-off error). However, by increasing the gamma, the newly generated pixels (the "dark" muddy pixels between the mauve and green) are gamma'ed out of existence --- they become very light gray. I would propose that a light-gray pixel will be interpreted as a transition zone in the light field your eye picks up, rather than an edge. The dark-gray transition in the light field will be interpreted as an edge. This makes the light-gray transition "look" better without having any discernibly different characteristics, other than being lighter-in-color.

(3) Upsampling Rasterized Vector Images
There are a number of (free) programs for converting from raster images (especially 2-bit raster images) to stroked vector images. I have had success with both POTrace and AutoTrace/Delineate. They can be found here:
http://potrace.sourceforge.net/
and here:
http://delineate.sourceforge.net/
respectively.

[RobA]

Another update on this topic.

There is great review of different up-sizing techniques over here: http://www.cambridgeincolour.com/tut...nlargement.htm

(as well as a wealth of digital photo tips, techniques and tutorial at that site!)

-Rob A>