>Thinking it through, I realized that arbitrarily sizing of a digital
>image could easily lead to strange results. If the dpi of the image at
>the set print size is not an even divisor of the dpi of the printer, the
>printer driver has to interpolate. Lets say it has to print 11 pixels
>from each 7 or 9 pixels in the source. There is no mathematical way to
>do that and maintain the integrity of the image! The software will do
>it's best, but the result will inevitably be poor. On the other hand,
>converting 10 to 5 pixels or 3 to 9, etc. isn't a big problem to do.
>
>So I tried letting the size of the image decide the actual size of the
>print. I made a chart of dpis that are even divisors of the 1400 dpi
>resolution of the printer and only printed at those dpis. So an 8x10 may
>turn out to be 7.7x10.2, depending on how it is cropped, etc.
I agree, and always do the same. Interpolation via the printer driver is
usually awful. However, Photoshop does a much better work, esp. with the
'Bicubical' option, but don't expect miracles!
>If you read the interchange between me and Joe Gwinn, you know some of
>the issues of grain (granularity) aliasing in digital scanning. It
>appears that, with certain films, upgrading from 2700 to 4000 dpi could
>actually give worse results!
Quite related -- if you take a 2700 dpi scan with a 4000 dpi scanner,
results will be worse than those of a "real" 2700 dpi scanner. I *always*
scan at the nominal dpi of the scanner. Then I try to get *close* to the
desired print size, downsampling by simple integer divisors of the
printer's resolution. When the picture *must* fit an exact size, first of
all I set that size in Photoshop and later resample to a "printer-friendly"
resolution with bicubical interpolation, but then the picture usually needs
a bit of sharpening.
>I've made really good looking 8x10 prints from cropped digital images of
>around 1000x1200 dpi, or about 1.2mp. That means to me that 5mp is good
>for 16x20 prints, as our E-10 and E-20 users have said.
YMMV. I like to print image files of 180 dpi (720/4) or more, but I usually
find 120 dpi (720/6) acceptable. That's 2.6 MP+ or 1.15 MP for an 8x10"
print! However, a fully-scanned, properly downsampled image of "x" MP looks
*much* better than a pic form a "x" MP digicam. If you compute the total
amount of pixels on the image (the "effective" pixel size), you'll get a
slightly different number than the CCD's pixels. This leads to the worst of
interpolations (that between similar resolutions) being done by the camera
itself! Maybe that Foveon-3D sensor (?) would improve this.
>Even better, how about
>greater dynamic range, greater sensitivity and less noise in the
>sensors, rather than piling on the megapixels?
Nice thing. I'd like to see digicams with *useable* high speeds (say ASA
1600 or so), and also low noise in "B" exposures. But I'm afraid we still
need some MP excess to make artifacts negligible.
...
Carlos Santisteban
<cjss@xxxxxxxxxxxxx>
<cjsantis@xxxxxxxx>
< This message was delivered via the Olympus Mailing List >
< For questions, mailto:owner-olympus@xxxxxxxxxxxxxxx >
< Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html >
|