In order to test various scaling methods it is helpful to decide on some meaningful tests, ones that are reasonably typical. The most likely scale-up factor is between two and four times, representing a nominal 12x8inch RAW file moving up to a 16x24 or, at the extreme, to 48x32 inches. Scaling any larger than this requires quite specialised large-format printers and assuming you have to take the printing to a bureau, they are likely to have their own RIP scaling or even GF6 as standard.
ABOVE: Scaling up and down is hugely damaging to a file regardless of how
the scaling is incremented. The image on the left is the original, scaled at
50% increments in the centre and nominally 35% on the right. Both files to
the right are damaged beyond use.
Scaling down is most typically from about 3,900x2,600 to 1,024x768, a
nominally three-fold reduction, for web use of image. Taking a picture down
even lower to say 800 pixels (quite common for web use) is a linear scaling
of 4.88 times.
Much of the older theory on scaling calls for incremental scaling, that is,
getting to a specific scale value in a number of jumps rather than a single
move. Scaling down to a nominal 500 pixel image may be accomplished by
scaling 3 times at 50% or twice at 35.36% to give the 8-fold linear
reduction. We also scaled directly down to the 484 pixel size and then
examined all the images. We could not tell the difference!
BELOW: The file to the left has been scaled once down to size, that on the right has been scaled many times to the same eventual size using 5% reduction steps each time.
Scaling down, then up again, is very bad practice, it devastates image
quality. We tried to move the files down 8-fold and the back up again using
eight moves and six moves. The files were badly damaged but neither method
was better than the other.
It is often recommended that scale down should be performed in very small
increments eg using a 95% value as much as 10 times. This is sometimes
accompanied by staged sharpening. Accordingly we wrote an action to carry
the process out and did more testing. Although the image had an apparent
higher sharpness over that reduced in a single move, the detail was rather
coarse.
Comparing a more modest 50% size reduction using GF6 and Bicubic sharper (in
Photoshop) revealed a detectable difference in the results. The GF6 was
smoother but the Bicubic was sharper and produced more detail in the puppy
hair along with more artefacting around the teeth. GF6 rendered the grass
more effectively.
ABOVE These screen grabs are made at 100% after the full size RAW file
was reduced to 50% linear using Bicubic Sharper and GF6.
Overall our conclusions were the same as those we have reached before. For
stock submissions you are likely to be faced with a demand to use GF so you
might as well get on with it. For really big poster enlargements GF6 gives a
slight edge and the feeling that you are doing your very best. For more
modest scaling Photoshop is more than adequate, particularly if you are
prepared to experiment and use incremental scaling, we have seen some
convincing demonstrations of the technique but it is probably image
dependent.
If you have to do a lot of image scaling and control over quality is
paramount, the batch processing (up- or down-scaling) of GF6 will avoid you
having to write and debug actions. The Pro version of the software has
additional features such as preparing files for canvas wrapping and working
CMYK files. It also operates in both Lightroom and Aperture as a plug-in.
Genuine Fractals 6 costs as follows:
Genuine Fractals Pro – £229.95 inc VAT (includes CMYK and Gallery Wrap)
Genuine Fractals Standard – £124.95 inc VAT
See www.ononesoftware.com
Read more articles on Travel photography here
