Stair Interpolation.... does it really work?

[mpenza]

http://www.dpreview.com/misc/rdr.asp?url=http://www.fredmiranda.com/SI/index.html

 

[mpenza]

tried it out....

original pic:

reduce the pic to 1/3 size and then use the SI algorithm to enlarge it to 3x.

remove noise using icolordenoise and unltrasharpen the above

 

[mpenza]

Enlarged 1.5x from original:

after sharpened:

seems not too bad

 

[yeppie99]

SI works quite well for certain cases. one good example: lets say i have a photo of a group of people but i only want one person's face to be printed out on say A4. if you crop in on that one person's face and blow it up to A4, its super pixellated, unless u have a very high MP camera. so use SI to enlarge the cropped face or enlarge the whole image then crop the face. much much les pixellation.

2 cents worth.

 

[mpenza]

yup. someone actually printed 20 inch x 30 inch from a 3.4MP camera.....

 

[yeppie99]

original

cropped and enlarged bicubically(new word)

cropped and enlarged using SI(less pixellation)

using Irfanview

i think the difference is more obvious from far

 

[yeppie99]

actually our humble 2 MP also can print so big, but remember to hang it high high or far far so people can't get near it.

 

[erwinx]

(1) theoretically, 1% stair interpolation is supposed to be better than simply bicubically resampling to the final size... its even officially recommended by adobe. Note that the free action provided is NOT 1% stair interpolation, but rather large jumps.

(2) On several photos i tried the Fred Mirandi SI action vs bicubic on, it seemed that SI produced less contrasty edges than standard bicubic. photo looks 'sharper' because more contrasty. However, it could be that 'more contrasty' = more jaggies when printed out...

i think the same thing is happening for the snake picture... look around the eye area, bicubic seems to produce clearer edges and is more contrasty.

(3) also note that Fred Miranda is comparing Genuine Fractals 2.0 which uses lossy compression. Genuine Fractals print pro is the pro version which uses lossless compression and may be better.



(as for the other picture, as its on homex, not viewable overseas )

 

[yeppie99]

so how do u go about using 1% SI? is there an action for it?

 

[StreetShooter]

Just create your own action lor.
Use Image Size command and set it to 101%.
I believe Fred uses 110%.

Bicubic has a touch of sharpening, I think, which is why the detail is preserved.

 

[kahheng]

SI works reasonably well. From my tests of it, it's generally better than GF. However, I prefer Irfanview, goto: http://www.irfanview.com and try out the various resampling algorithms in this freeware. (specif: Lanczos, Mitchell, B-spline)

I've gotten great 20x30 inch prints from a 3.3 MP camera with Irfanview. And it's FREEWARE still.

FWIW, diff algorithms work well for different kinds of scenes. It's impossible to have a one size fits all solution for all types of pictures.:gbounce: :gbounce: :gbounce:

 

[kahheng]

Originally posted by erwinx
(1) theoretically, 1% stair interpolation is supposed to be better than simply bicubically resampling to the final size... its even officially recommended by adobe. Note that the free action provided is NOT 1% stair interpolation, but rather large jumps.

(2) On several photos i tried the Fred Mirandi SI action vs bicubic on, it seemed that SI produced less contrasty edges than standard bicubic. photo looks 'sharper' because more contrasty. However, it could be that 'more contrasty' = more jaggies when printed out...

i think the same thing is happening for the snake picture... look around the eye area, bicubic seems to produce clearer edges and is more contrasty.

(3) also note that Fred Miranda is comparing Genuine Fractals 2.0 which uses lossy compression. Genuine Fractals print pro is the pro version which uses lossless compression and may be better.



(as for the other picture, as its on homex, not viewable overseas )

Erwinx, GF 2.0 gives you the option of lossless fractal encoding as well. You select it when you encode, and decode. I have been using GF since v1.x days and it's allowed this for the longest time. The Print Pro version merely allows you to encode CMYK files. If you don't use CMYK files, it's pointless.:rbounce:

 

[erwinx]

Hi KH,

I downloaded a demo version of GF and it clearly doesn't allow lossless compression. I try to save at the max quality the dialogue allows and the file sizes are far too small (compared to compressed tif) to be lossless.

I guess it doesn't apply to the full version... (it didn't say anything in the help file about this)

Originally posted by kahheng



Erwinx, GF 2.0 gives you the option of lossless fractal encoding as well. You select it when you encode, and decode. I have been using GF since v1.x days and it's allowed this for the longest time. The Print Pro version merely allows you to encode CMYK files. If you don't use CMYK files, it's pointless.:rbounce:

 

[kahheng]

Erwin

I think you might have mistaken fractal encoding with JPEG compression. It's a different thing altogether.

If you chose the "Lossless" option for encoding, none of the image data would be destroyed. theoretically.

If you think about it, TIFF can also be LZW compressed, losslessly.

 

[erwinx]

there is no lossless compression option in the GF demo i downloaded, only 'Pro graphics' and 'web graphics'. Both are lossy. The reason i say this is that the file sizes are about the same as the file sizes of the original Jpegs. It may well be that I downloaded a very old version of the demo or something.... (i followed some links and didn't download from altamira's site) the copyright notice says 1996

Originally posted by kahheng
Erwin

I think you might have mistaken fractal encoding with JPEG compression. It's a different thing altogether.

If you chose the "Lossless" option for encoding, none of the image data would be destroyed. theoretically.

If you think about it, TIFF can also be LZW compressed, losslessly.

 

[yeppie99]

Originally posted by StreetShooter
Just create your own action lor.
Use Image Size command and set it to 101%.
I believe Fred uses 110%.

Bicubic has a touch of sharpening, I think, which is why the detail is preserved.

can you describe how to go about creating such an action? i don't want to go to Image Size and change it 100 times before I get about 200%. Not so familiar with PS actions. I can create an action that resamples it 101% but how do I duplicate it?

 

[StreetShooter]

This should be self-explanatory.
After that, you just keep hitting F2 until the picture reaches the size you like lor.

You can, of course, create different percentage increases eg 5%, etc and assign them to different function keys.

 

[yeppie99]

ok thanks. but still no way to automate the process, say perform this 1% or 5% action 50 times?

 

[StreetShooter]

Can. Once you have created the action, you can see the individual commands listed under it in the action panel. Just drag that command to the "New Action" button, and the command will be duplicated. Do this as many times as you want. Then, when you run the action, it will do the same thing, say, 50 times.

 

[Jed]

Actually, there are no real facts to show that interpolating in small steps produces a better results that upsizing in one step, assuming the interpolating program is a good one. It's been tested in a trade magazine that I respect a lot that interpolating a clean diagonal in small steps actually introduces weird artifacts to the resulting image. From the interpolated lizard heads, it does seem like step interpolation works better, but I don't think the bicubic in one step was done with the best algorithm because I'm sure I've seen much better.

Personally, I use GF and it works a laugh, but for most situations, I don't need to hit that far. Photoshop's bicubic does a good job, and frankly most of the time I don't interpolate at all.