About the real resolution of flatbed scanners

[lotas]

Hi

According to what i've read in this news group, the effective
resolution of flatbeds tends to be less than a half of that specified
by the manufacturer.

Now,here goes my question: Let's take Canon's 9950f ( 4800 x 9600 dpi
). If it's real dpi is - just guessing - 2400 dpi, does it means i can
scan negatives at 4800 and then downsample to 2400 without loosing
quality at all?

regards

 

[Kennedy McEwen]

Hi

According to what i've read in this news group, the effective
resolution of flatbeds tends to be less than a half of that specified
by the manufacturer.

Now,here goes my question: Let's take Canon's 9950f ( 4800 x 9600 dpi
). If it's real dpi is - just guessing - 2400 dpi, does it means i can
scan negatives at 4800 and then downsample to 2400 without loosing
quality at all?

Without losing quality at all? Nope.
Without losing significant quality? Probably, in fact almost certainly.

 

[Leonard Evens]

Hi

According to what i've read in this news group, the effective
resolution of flatbeds tends to be less than a half of that specified
by the manufacturer.

Now,here goes my question: Let's take Canon's 9950f ( 4800 x 9600 dpi
). If it's real dpi is - just guessing - 2400 dpi, does it means i can
scan negatives at 4800 and then downsample to 2400 without loosing
quality at all?

It is highly likely you won't lose any quality that you can detect in
any practical sense.

I'm not familiar with the Canon, but I regularly downsample my Epson
3200 scans because otherwise the files are hard to store and it takes
forever to edit them in a photoeditor. From time to time I comapre
before and after downsampling, and if I see a difference it is so
slight that I am not sure it is just my imagination. Most of the time I
see no difference whatsoever.

By the way, you should be clear about one thing. Calling it "real dpi"
or "real ppi" is misleading. The scanner does deliver the number of
pixels per inch that it advertises. The problem is that it doesn't
resolve photographic detail at that level. That is, the extra pixels
are in some sense empty pixels because they don't exhibit the desired
detail which the advertised dpi or ppi is theoretically capable of.

 

[Tim]

Yes, quite correct, the effective resolution of any flatbed scanner is
about half the number of pixels. The resolution is measured as the
number of pixels.
To scan any line the sampling needs to be half the diameter of the
line. Think - when the sampling resolution is the width of the line
consider a sample that contains half line and half gap between the
lines. The next sample will include the other half of the gap and half
the next line. As there can only be one RGB value for each sample how
can you expect the scanner to resolve both the line and the gap?
Now repeat the exersize with the sample as half the width of the line.
You will now always have one sample fully on the line and one sample
fully off the line...

 

[Leonard Evens]

Yes, quite correct, the effective resolution of any flatbed scanner is
about half the number of pixels. The resolution is measured as the
number of pixels.

The theoretical maximal resolution in line pairs per unit length is half
the sampling frequency in pixels per unit length. This has nothing to
do with whether it is flatbed scanner or film scanner.

The problem is that no scanner will ever achieve its theoretical
maximum, although some will come closer than others. Flatbed scanners
tend not to do as well as dedicated film scanners in this respect. The
Epson 3200 delivers at most 30 lp/mm in photographic resolution. This
has been verified by various people doing their own tests. Its sampling
resolution is about 126 pixels per mm. The theoretical limit would then
be about 63 lp/mm, and what is delivers is at best half of that or one
quarter of the sampling resolution.


This makes the Epson 3200 adequate for much medium format photography
and more than adequate for 4 x 5 photography. But it is generally
inadequate for 35 mm photography. The Epson 4870 apparently does a bit
better but not 4800/3200= 3/2 better.