Upresing and Noise.

[Joe Copalman]

I was recently contacted regarding the use of one of my photos for the cover of a scientific journal.  The problem is that the photo in question was taken back in 2006 with a 5MP point-and-shoot and making the image work at the requested resolution is turning out to be a headache.

The native resolution is 180dpi with the proportions being 10.8' wide x 14.4' high.  Corresponding pixel count is 1944 x 2592.  The resolution being requested is 8.625' x 11.375' @ 300dpi, requiring me to upres the image.  The original image is pretty noisy at full-size, so I applied some noise reduction through PS Elements 6.0, before and after re-sizing, and after experiment with several combinations of strength and detail preservation, I still cannot adequately control the noise as much as I would like.  Sharpness of the main subject - a saguaro silhouetted against a sunset - is slightly problematic as well, but not as bad as the noise in the blue-sky portions of the image.

I have very little experience in upresing photos and am trying to give myself a crash course by reading various articles on it, but any input from anyone with more experience would be GREATLY appreciated. 

[Jay Beckman]

I don't know if the free trial inserts a watermark or not but Perfect Resize from OnOne is pretty much the benchmark software for up-resing.

You may know it by its previous trade name: Genuine Fractals.

[Joe Copalman]

Thanks Jay!  I requested the download link for the free version, I'll download it tonight.

I ran the photo through Neat Image last night and am beyond impressed with the results.  Did it prior to resizing and it killed a lot of the noise while keeping most of the detail.  It did result in a handful of tiny artifacts (red or white dots about 4 pixels in diameter), but those were a breeze to clone out).  Resulting image looks pretty good, especially when viewed at the size it will actually be printed at.

[Jay Beckman]

Should you ever decide to change up your workflow and migrate to Adobe Lightroom, you'll find that it scales image very well.

I think it was Luminous Landscape did a comparison of several pieces of software for scaling and IIRC, Lightroom fared surprisingly well Vs the stand along / plug in products.  It might have even come in at #3 behind Fractals and (I forget which...)

I guess I've never really stopped to think about it, but seeing as everything that goes up to the commercial sections of my website is 5000 on the longest side, quite a bit of my stuff is being enlarged at least a little and the prints I have that have come from MPix don't really show any artifacts or jaggies at all.

[Joe Copalman]

Yeah, the one shot I've printed from Mpix so far was up-resed to 5000x per your suggestion and it came out great at 12x18.  And that was just using PS Elements to resize it. 

Lightroom is in the future, it's just number five or six on a list of things I'm counting my pennies for at the moment.

[Jay Beckman]

Found this chart that helps explain Megapixels Vs Optimal Print Resolution and Size

http://www.design215.com/toolbox/megapixels.php

[Dave S.]

One of the 'tricks' I read about to get a good up-size was to first double the image's dimensions.  After doubling it, you then crop/re-size back down to your final, desired size.  By doubling (or tripling and even quadrupling - basically using any whole number, within reason) the size - and this is the key point - each and every pixel is affected the same way, in a symmetrical fashion.

For example, starting with a pixel width of 1024 pixels and you wanted/needed to print it at 300 ppi & 10 inches = 3000 pixels.  I would look at two approaches to get to my final, 3000 pixel dimension:

If you were to up-size the image straight to your desired size, it may be an increase of, say, 292.9% (i.e multiply 1024 x 2.929 = almost 3000).  If this happens, you then rely on the computer to interpolate when, where and how to split your 1024 pixel width dimension into 3000 pixels, not a nice, clean solution.

But, if you were to quadruple your image first (1024 pixels x 4 = 4096 pixels), each pixel is split into four evenly sized pixels, along each edge.  Now re-size the image back down to your desired 3000 pixel dimension.  This results in only a ~26.7% reduction in size.  To me, this method seems a bit cleaner from a computer/mathematical sense, and you end up minimizing how much guesswork the computer/software has to do (i.e. a small change of 26.7% vs. the leap of 292.9%) and I would imagine you'd end up with fewer artifacts with that smaller change).

Keep in mind that as you upsize your image, there is a geometric relationship that effects your image.  Each time you double the image's size (i.e. go from 1024 pixels to 2048 pixels), you're going to increase the overall file size by a factor of four.  That's because you will also be doubling the other dimension in the image at the same time.  Since we're effectively dealing with the surface area, just like in high school geometry - if you double the size of a rectangle's dimensions (i.e. go from a 5x7 to a 10x14), you're going to increase the area by a factor of four (5x7=35, 10x14=140 & 140/35=4).  In the example above, if you quadruple the long edge of your image (1024 x 4 = 4096) you'll also be quadrupling the short edge of your image (say 623 x 4 = 2492).  With the geomtric relationship, this should increase our image size by a factor of 16!  Now, 1024 x 623 = 637,952.  Then it stands to reason that 16 x 637,952 = 10,207,232 which also happens to = 4096 x 2492, thereby demonstrating our factor of 16.

The above in purely a mathematical examination and does not take into account any algorithmic approaches that some of the better software can/do incorporate when increasing an image's size.  Nonetheless, I may just play with this and print each method, and then compare the results.