Thanks Bart for the additional information and advise. As is so often with these things, one needs to try it out for oneself. I went ahead and downloaded the 14 day trial for Qimage Ultimate, and proceeded to make some comparisons and prints. Indeed, Fusion, Hybrid and Hybrid SE are all significantly softer than PZP or Resize on close inspection, but for small amounts of upsampling, the visual differences in print are small. By the same, the differences between Qimage's interpolators and Photoshop's bicubic are small too. I am interested in getting the best quality out of my images, 'almost there' just isn't good enough.
Hi Samuel,
Yes, the differences may be subtle at times, but a lot also depends on the subject matter. Sometimes the differences are more obvious than subtle. Of course, PZP and Perfect Resize can create even better upsampled result, as can Topaz Detail for detail enhancement, but these are specific image processing applications/plugins, not print management systems, and the results are (often large size) one-of's in a more involved workflow.
Mike Chaney has also said something to that effect, that Qimage may not produce the
absolute best possible output, but produces by far the most convenient/consistent/repeatable high quality output hat usually leaves e.g. straight forward printing from Photoshop or other applications in the dust. This is often confirmed by new users of Qimage, they can clearly see the improvement compared to their e.g. Photoshop output. And in addition, QU adds a lot of print management workflow tools, including recalling prior printjobs and unclogging of single color channels, to name a couple.
Regarding the speed of Qimage, I am left befuddled. The trial version is x86, so I get memory issues when trying to work on large files to test Qimage's powers with huge data volume. But even with relatively small file size (the test file I used was 7400x9516), saving out a TIFF took 53 seconds and DFS took 56 seconds to run.
Could you explain what was done exactly? When you say 'saving out a TIFF', is that upsampled? Also remember that 16-bit/channel data (and layers) is not absolutely necessary if you produce printed output. Qimage will (dither and) round to 8-b/ch anyway, to accommodate the printer driver. The dithering helps to visually smooth out potential quantization effects like posterization. You can also feed a highest-quality JPEG (with print-medium profile or a large colorspace) to Qimage, you'll be hard-pressed to see any difference in print. Qimage will do the resampling, sharpening, optional dithering, and finally the conversion to the output colorspace anyway.
And how did you determine the time DFS added (e.g. printed without and printed with DFS?)? Did you print to file, in which case Hard disk performance and Virus scanners can make a difference. Also make sure that you have (Edit>Preferences) Multithreading (Multi-Core processing) set to whatever works best on your system, and that Qimage isn't still building thumbnails in the background. Also memory fragmentation can play a role, which can be checked with holding down the Shift key when you click 'Help>Analyze Current Settings'. If the print file is larger than the largest contiguous memory block, then the Operating system has to manage memory, and maybe swap to disk. Remember that Print output files can become rather large, despite their single layer 8-b/ch data.
In comparison, Photoshop saves the same file in less than 1 second, and Topaz Detail took less than 2 seconds after hitting the OK button (not counting the pre-processing at the initial launch stage). I don't know if the paid version is any better in this regard. If it isn't, it would be very frustrating for me to use.
Again, I do not know what we are comparing, so it's hard to comment on that. Doesn't Photoshop's Autosave play a role in this? And is the save time with or without resampling time? Qimage does everything at the same time, Resampling/Sharpening/Profile conversion/Dithering/anti-aliasing if down-sampling, and sending chunks to the Printer driver to avoid memory issues. I'm not so sure that there is less efficiency if you add up all the steps required to produce output, but maybe there is room for improvement, who knows? I'm sure Mike Chaney is willing to listen.
I took a good look at DFS. Actually DFS does produce halos, and they are quite wide and quite ugly. They are not obvious at 'normal' light handed settings, but the halos become very strong at high % settings, even when the radius is set to 1. I can easily make out the halos for Radius 1, 200% on screen, and I can just barely see the halos in a 600ppi print (image upsampled 400% by Fusion).
That doesn't correspond with my tests, although we might interpret halos differently. Also make sure to view DFS separate from Resampling, which usually
will produce a certain amount of halo (only something like a low contrast bi-linear resampling can avoid that, or special procedures like PZP uses).
If we take an abrupt change in image brightness between a uniform dark gray and a uniform light gray region, then sharpening that must enhance the edge contrast by lowering the dark gray values, and raising the light gray areas, otherwise it would not look 'sharper'. That's not really a halo, it's enhanced acutance, or edge contrast, there is no other way to give more acutance to such a transition.
On a slightly more gradual transition (1 or more pixels) from darker to lighter one can use low amounts to reduce over/under-shoots , but they will not be very visually effective, and larger amounts may exaggerate the edge transition, but again that is simply the result of boosting acutance with a USM type of sharpening. DFS effectively suppresses that, but cannot (should not) eliminate the acutance boost, it just avoids the additional halo.
DFS has the tendency to lower the contrast of already low contrast detail. However, it is very good at avoiding sharpening and enhancing noise, which leads me to wonder if there is some kind of noise reduction going on too, which may explain why low contrast detail gets flattened. Fine textures are also smeared out by DFS, which results in a blotchy, plastic looking image. But - the edges of mid-high contrast detail is enhanced in a way that Topaz Detail cannot do. DFS does appear visually sharper in print for fine detail like twigs against sky albeit with some slight halos, but on textured surfaces like tree bark, it does not look as natural as the Topaz Detail rendering.
Topaz detail is rather unique! But it is not a print mamagement application.
Topaz Detail is a totally different kind of sharpening. It seems to be more like a texture enhancement tool than a sharpening tool (one has to be careful about noise).
Correct, besides the 'Deblur' slider control which performs deconvolution sharpening, it halo-free boosts spatial frequency detail. A negative Boost amount will reduce the amplification of graininess.
It's more like Clarity but with a much smaller radius setting (I almost exclusively use small details settings in Topaz), and without the significant lightening or darkening of details on each side of the edge. Topaz Detail truly adds no halos even when the 'Small Details' setting is all the way up at 1.00.
Yes, but 'TL Clarity' boosts contrast, while 'TL Detail' boosts spatial frequencies.
I've long known about Qimage and have read many many good things about it. I'm just not sure if it really provides a leap of quality for my work. I hold anything you say in high regard, of course. Many recommendations and tips you have so generously provided have helped me tremendously to improve. I'm just not seeing this in Qimage at the moment.
It's hard to compare hand tweaked results achieved by using a number of hand picked best-of-class procedures, with largely automatic productivity tools. The workflow benefits cannot be easily compared with hand tweaked results, because we are comparing different features and do so while pixel-peeping. In real output the differences will be hard to detect though, not impossible, but hard.
Cheers,
Bart
