2 Replies Latest reply on Aug 30, 2011 10:39 AM by Dov Isaacs

    Confusion about upsampling?

    John Hawkinson Level 5

      I'm not sure this is really the optimal forum for thus, but it seems about as good as any.


      In a few recent threads, including Completely Embed Font (not Subset) in PDF several methods not working, Dov
      Isaacs has said things like:

      Don't try to artificially upsample or downsample any image before placing into InDesign. Upsampling does not improve quality in any way whatsoever.

      (I'm posting this seperately because I don't want to confuse those other threads. And Dov's advice is so generally good.)

      I'm finding this claim about upsampling to be rather confusing, as it is counter to my understanding.

      I'm under the impression that upsampling can indeed help to interpolate points and lead to apparently sharper images and finger detail. While it cannot restore data that is never there, it can be a marked improvement over not doing so.


      As a practical example, we screen at 220lpi and print at 1200dpi, and one of our regular sources of cartoons (line art) comes in at 72ppi (about 110ppi effective), despite our best efforst to get higher resolution. We use a Photoshop action to upsample it to 300ppi (about 360ppi effective).


      Specifically, we start with a 72ppi/8pp anti-aliased grayscale image, we upsample it to 600ppi, threshold black at 128, convert it to a bitmap at 600ppi with a diffusion dither, and downsample to 300ppi/1bpp.


      The result looks dramatically better in print than the original. Is this a bizarre corner case (because it is hand-drawn line art; because the original is anti-aliased; because we're dithering; because this starts to sound like "image processing" rather than just "upsampling")? Because I would tend to think it is not consistent with the idea that upsampling doesn't help.


      I am less sanguine about the next part of this: on occassion, some of our photo editors have had images they want to run in print whose resolution is much much too low to look acceptable. But for which the image is fairly important to run, regardless. They had had non-negative success with upsampling in Photoshop and perhaps applying some blurs. This can result in images that don't have the telltale artifacts ("huge jaggies") of low resolution imagery, even though they may not really have more information density. But there is value in eliminating those extremely obvious artifacts,

      even if they are ultimately replaced with other artifacts.


      So, am I missing something? Am I just confused? Or do these just qualify as corner-cases?



        • 1. Re: Confusion about upsampling?
          Dov Isaacs Adobe Employee

          Like all things graphical and technical, there are always corner cases.


          In terms of upsampling, every RIP and every renderer out there upsamples images when necessary, i.e., when the effective resolution of the raster image (native resolution tempered by the magnfication for print or display) is less than the resolution of device on which they are being printed or displayed. And the converse is true as well in terms of downsampling.


          In fact, with the exception of one-bit-per pixel monochrome images at exact device resolution, all imagery is resampled during the process of halftoning. You are not avoiding a resampling process by trying to hit some magic number in terms of resolution!!


          What you do want to avoid like the plague is the result of cascading resampling of imagery. All raster image upsampling operations are at best speculative in terms of how they “invent” the extra pixels for interpolation. What is used in Adobe Photoshop as well as by other Adobe applications including Acrobat and the Adobe PDF Print Engine RIP technology by default, bicubic interpolation, is generally very effective and high quality. All raster image downsampling is lossy. And if you upsample (interpolate) an image and then downsample it, which pixels do you lose, the original information or the invented information? The results can be quite nasty.


          The best example of this problem is that of screen shots. What is the best way to reproduce a screen shot? You will get all type of opinions on this, but what you definitely never should do is to upsample it. (There are some interesting techniques that can be applied to clean up black text in such screen shots, but they don't involve attempts to change resolution.) Simply place the screen shot “as is” and let the screen display software in your application or in Acrobat or the RIP for printing resample the image once during the screening process.


          Note that there are various Photoshop plug-in packages that claim miraculous results for upsampling requirements, but these miraculous results are typically representative of situations in which really extreme upsampling is required and not the garden variety that most of us would normally deal with. Of course YMMV.


          You also need to be exceptionally careful about any resampling, downsampling or upsampling given the havoc it can cause with images that have had any significant sharpening already applied to them.


          Hopefully this cut the confusion a bit ...     These issues are probably one of primary reasons one should stay with text rendered via fonts and vector graphics as much as possible before resorting to use of raster images. (BTW, one of these days the geniuses at Microsoft and Apple will learn how to provide a screen shot facility that uses text and vectors in addition to rasters based upon the original display lists provided to the underlying windowing system!)


                    - Dov

          • 2. Re: Confusion about upsampling?
            John Hawkinson Level 5

            Well, I guess the upshot is I'd be careful saying there's no quality increase to upsampling, since, at least to me, line art does not feel like a corner case.

            (Maybe that was key here was the thresholding operation, i.e. the elimination of the poor interaction between antialiasing and halftoning.)


            If I can't get this artwork at 300ppi  and am stuck with getting it at 72ppi, there's pretty much no way I am going to get it in vector form! Much as we'd love it.