Hi Bart,
I'd like to hear your thoughts on the matter of up-sampling. Currently the best I can find is the 'hybrid' method in Qimage, but I suspect there may be better.
It will of course be difficult to judge which is best without printing, because as Rachel discovered all severely up-sampled images look awful when viewed on a monitor.
Hi Stuart,
As usual, it depends on the enduse of the image. In general, it doesn't matter how a close-up looks (on screen), it only matters how it looks at the intended output viewing distance/magnification. Another criterion that may be important is how long it takes to produce an enlargement. For most purposes, it would not be acceptable if it took several hours per image to produce an enlargement.
Having said that, there are a couple of methods that have a foundation in scientific research and that have a reputation of producing a good compromise between the possible artifacts associated with interpolation. Subject matter, e.g. line drawings instead of photographic images, can play a role in choosing the best compromise.
In general, an algorithm known as 'Mitchell' (from the researchers Mitchell and Netravali) strikes a good compromise when upsampling/interpolating. It retains some of the sharpness, but avoids severe blocking, ringing, and aliasing. It is a specific mix of parameters used in a more general cubic filter algorithm.
Qimage, with an emphasis on print output, offers a very efficient workflow and it delivers high quality output. It also offers the Mitchell type of interpolation, but also some other methods. I prefer the Hybrid SE method (only available in the Studio Edition) for general work, because it delivers very natural enlargements with an even better suppression of jagged edges than the Mitchell method does, but at the expense of a little sharpness (there is always a trade-off). The Hybrid (i.e. non SE) method adds a little edge contrast which might turn into visible halos, but only on very large magnifications. Hybrid SE offers a very good startingpoint for further sharpening, e.g. time consuming deconvolution sharpening, when one prints to a file instead of a printer.
There are also other applications/plugins that can help with enlarging, but they require input for the best results, and are not always the best solution for a specific task. For the best results, one should have a good understanding of the specific goal and output modality. Qimage offers a very good path to quality output, but the interface takes a bit of getting used to. For most photographers who produce printed output, it's a no-brainer, a must have program for the toolkit. It is a Windows program, but it apparently also runs fine on Macs (e.g. with Parallels).
Cheers,
Bart