Let's say I have a file of 2880 x 3600 pixels and want to have a 16x20 print. This gives us a resolution of 180 ppi, generally accepted as a minimum for a decent printout. Alternatively, I can interpolate the file to 4800 x 6000 pixels and print it at 300 ppi, a native resolutions for many printers. Which method would give a better-looking result?

I know that interpolation does not add any new detail, the same way sharpening also does not add any new detail. Yet, at least when sharpening is concerned, the end results appears sharper. So, the question is: would interpolating the file to 300 ppi produce an appearance of a higher quality print? Has anyone done such a test?

My understanding is that the printer driver will interpolate the file to the native resolution of the printer, in my case to 360 dpi for epson large format. Unfortunately it won't use the most sophisticated method and even bicubic in photoshop will do a better job with less artifacts.

If you look at the QImage website (ddisoftware.com) I believe they have a more detailed explanation as well as some test images to demonstrate on your own printer. Obviously the intent is to sell QImage but I think they make a good case for taking the up/down rezzing away from the printer driver.

It's been awhile since I've done testing myself but at the time it could be pretty convincing, especially if the distance between the native printer resolution and image was great and not near an evenly divisible number (I.E. not 180, 240, etc but more like 217, 276 etc). I would check to make sure of the printer driver's resolution and not assume 300 dpi.

I use QImage for most things but usually rez-up to the driver resolution even printing from photoshop.

Good luck....
Doug

According to the chief printing scientist at HP, whom I interviewed about 9 months ago, printer drivers do a better job of interpolation that does anything else between the range of 180 to 480 ppi. In other words, if your image is in that range leave it alone and just feed it to the printer.

If it's below 180 use Photoshop's Bicubic smoother to res it up, and above 480 user Photoshop Bicubic Sharper to res it down.

I've tested this extensively myself with Epson, HP and Canon large format printers and found it to be the case.

When ressing up I usually go to 240 and when ressing down I usually go to 360.

Michael

Michael,

I have doubts that HP engineers have used Qimage as an application to print from. The general statement that printer drivers do a better job than anything else on extrapolation isn't justified if they only use PS. The driver is also a black box while Qimage's proof print will show you fast what extrapolation type and degree is suitable and how print sharpening works out best. All without changing the image file as it does it on the fly. I have yet to see good anti-aliasing done in down sampling in a printer driver.
There are reasons to think that HP did improve the Z3100 driver's routines but on my questions to improve the cooperation between the driver and Qimage I never got an answer so I doubt they ever used Qimage.


Ernst Dinkla

try: http://groups.yahoo.com/group/Wide_Inkjet_Printers/

Perhaps that is because it won't run on a Mac, and Qimage remains in the dark ages....

That's just silly.

Perhaps, there is no Qimage information in the LL pages either but Steve's Digicam covers it well and DPreview mentions all major upgrades. 6000 Qimage mailing list members can not be wrong and the user's number must be much higher. Wonder where the light shines ...



Ernst Dinkla

try: http://groups.yahoo.com/group/Wide_Inkjet_Printers

For me the takeaway from this discussion and others like it is that we're maybe talking about a distinction without a difference at this point. I've never done any side-by-side analytical tests, but I have printed straight through drivers, through Q first, and with high-end rips. They are all pretty good.

totally agree

I did a side by side a few weeks back wondering about the same thing.
I used an image of a lady holding a cat that was captured using a canon 1dsmk2. I printed it on an HP z3100 from a 300 dpi file then I downsampled in photoshop using bicubic to 180 dpi. The print at 300 was perfect as far as I could tell but in the 180 dpi file one cat whisker ( I shot it with the 135mm lens at f2.2) that was perfectly in the focal plane had jaggies that made it looked like it had little barber pole stripes. The jaggies were not visible on the monitor at 100%.
I'll always upsample to 300 if the image has fine detail. I use genuine fractals to upsample anything more than 130%

Charlie

No doubt the more detail you start with the more you can ouput, but once you are down to that 180 dpi, which interpolation method will yield the best printed result? (Just a rhetorical question, not pushing any agenda.)

Hi,
I used to put a lot of work into getting good big prints from my old canon 1D 4.1 Mpix camera and have a few suggestions for you. While the printer drivers might be the best at upresing, you might find some more pleasing results using one of the many programs to upscale digital files.

What algorithm will work best kind of depends on what your image has as its subject. I've found that lanzos (geniune fractals) type routines do best with inorganic objects such as bulidings, boxes and stuff, while spline and other mathematical routines do better with plants, flowers, people. Lanzos will upsize a dot into a square, while s-spline will make a square a circle when it upscales. Circles look more natural to me, but YMMV.

You'll have to pay particular attention to sharpening. I like to do this on masks (and use pixel genius photokit sharpener) You do want to sharpen some things before upresing like certain edges you want to be really defined, and other things not. You kind of have to understand what how the upscale software is going to work, and that takes experience. Sometimes you'll get ugly halos if you sharpen too much in advance and other times you need some before upsizing to get the edges to hold. Again it depends on the subject, and you'll have to sharpen again after upresing, both for edges and again for print output. It takes a lot of work to really make a good print from a small file. Oh yeah, if you have a noisy file you need to run neat image or some other noise removal program to clean up the file. Noise gets amplified in upscaling.

If this seems like a lot of work it is. One thing to consider is lightjet output from a service bureau. The lightjet print engine has a very good upres routine that includes sharpening. I've gotten great results this way without as much effort. I don't have the HP but have tried this with both the epson 7600 and 4000 printers and found that by upresing to 360 dpi using smart algorithms and careful sharpening and noise reduction, I could get a significantly better looking print than just sending 180 dpi.

Anyway good luck!


btw - I have found that I've gotten the best results with PhotoZoom Pro (which used to be S-Spline pro) as it has many different mathematical upres routines.

The different opinions probably have much to do with different workflows anyway. The people that try to get as much data in their takes as possible (scan backs, 4x5 scans, MF digital backs, FF sensors) will see less of extrapolation artefacts with whatever application or driver. If they deliver the printer its native resolution right away 1:1 300-600-1200 PPI for the HPs and Canons and 360-720 PPI for the Epsons (perfectly shown above Qimage's preview window) then there is no extrapolation done on the image in the driver and probably little in the application with that quantity of data. The remaining risk probably is in downsampling if the application isn't good in anti-aliasing, another filter that is subject specific.

http://www.xs4all.nl/~bvdwolf/main/foto/do...le/example1.htm
http://www.xs4all.nl/~bvdwolf/main/foto/do...down_sample.htm

The majority of Qimage users and most likely the majority of digital camera users are faced with upsampling as the only affordable way to make larger prints. If they feed the file straight away then it depends on the application whether it will upsample the image to the native resolution if the driver asks for that or just sends that data to the driver to let the driver do the upsampling. Qimage belongs to the first category in default mode but can be set to deliver the data straight to the driver. The last is nice if you rasterised PDF/EPS vector files in Photoshop with anti-aliasing on and the PPI set to 600 for a Z3100, both Qimage and the driver will not influence your fonts and vector edges then. With photography the default extrapolation and sharpening routines or other settings more specific for the content will do a better job. Qimage is always in competition with other specialised extrapolation programs on extrapolation routines and there will be few ahead of it in that game. Image content plays a role there too in what application is best. None of the specialised programs have the other print features of Qimage and none including a less specialised program like PS does handle it so conveniently on the fly without changing the image file like Qimage does it. That doesn't mean the user shouldn't know the basic facts of extrapolation like less or no smart print sharpening in real blow ups. The proof print feature where you can see the effects on a smaller piece of paper is a fast learning course as the image in the program isn't affected while you have access to several routines for the next proof. On a 1.5 x 1 meter blow up of a 20D 8 Mp file it is no shame to make two small crop proofs on the same scale. There will be extrapolation artefacts in that print, the question is which application or driver does it better. Including the Wasatch SoftRip that I have, the Scanvec Amiable and the Onyx RIP a friend has. On Epsons, Canons and HPs that we both have. Right now we both go for Qimage when requests like that land on our desk. The other question, which one does it faster has one answer: none.

Mike Chaney gives better explanations though the extrapolation article is already behind the latest Qimage upsampling routines.

http://www.steves-digicams.com/techcorner/tc_index.html


Ernst Dinkla

try: http://groups.yahoo.com/group/Wide_Inkjet_Printers/

When you give too much resolution to the printer and let the printer driver do the downsampling (e.g. giving 600ppi's worth of data) I'm guessing (and I mean guessing - this just popped into my head as an interesting theory) that the printer driver can do a good job of it because it can handle the dreaded jaggies because it can do clever things with the dithering. It has lots of available dots with which to render a single pixel, whereas a monitor only has one pixel to render a pixel and similarly if you downsize the original file you can only represent a pixel with a pixel.
Does this make sense or am I smoking something?
Anthony

Michael, when printing thru ImagePrint, is this still the case? Does IP use the native drivers, or its own proprietary driver for the printer? I'm printing on the Epson 4000, btw.