Epson 4870 Test

[Mike Engles]

Hello

I am very happy with this scanner, but in the interest of gaining more
knowledge about its optical resolution, I am considering the following
test.

I have a lens test chart that has various rulings.

I thought of scanning the finest ruling at 2400, 3200 and 4800 DPI.

In Photoshop I would resise the 2400 and 3200 scans to be exactly the
same size in pixels as the 4800 scan.

I would take the resized 2400 scan and layer it above the 4800 scan,
align it as best I can to make them register,difference them, flatten
them and apply Image equalise.

I would do the same for the resized 3200 and 4800.


Would this be a valid way of seeing if the Epson has a optical
resolution better than 2400?

If it were valid, would it be better to scan at a 45 degree angle to the
scanner bed?

Mike Engles

 

[-]

In Photoshop I would resise the 2400 and 3200 scans to be exactly the

same size in pixels as the 4800 scan.

I would take the resized 2400 scan and layer it above the 4800 scan,
align it as best I can to make them register,difference them, flatten
them and apply Image equalise.

If you are talking about interpolation, don't do it. As I suggested
earlier, open all three images side by side in Photoshop. Show the lower
dpi 2400 scan at "actual pixels" or 100%. Now use the Navigator's
zoom/reduction function to decrease the magnification until the higher dpi
images are the same size on screen as 2400 ppi image. What you do NOT want
to do is start enlarging files above 100%, or actual pixels. This will give
you a good way to do a visual comparison of the three different scans at one
time with minimal distortion by you computer. It isn't perfect, but I
believe it is better than interpolating files.

According to my calculations against the test images I posted a few days ago
that masked_off of unused areas of the holder and the scanner's glass, my
4870 produced a resolution of 31 lp/mm at 3200 ppi (which is actually
slightly better then my Epson 3200 at 3200 ppi) and 39-40 lp/mm at 4800 ppi.
There is a bit of interpretation/judgment call involved here but I tried to
be conservative when a judgment call needed to be made.

If you haven't tried a comparison between no makings and then masking off
the of the unused areas OEM Epson holders and scanner glass, try it. Make
sure to use a flat black material that is as light absorbing as possible (no
glossy or white). You should be able to clearly see a slight improvement.

Doug (the other Doug)

 

[Leonard Evens]

Hello

I am very happy with this scanner, but in the interest of gaining more
knowledge about its optical resolution, I am considering the following
test.

I have a lens test chart that has various rulings.

I think you are making it too complicated. This is the way I've done
it. Scan a test chart with resolution bar patterns at the highest
optical resolution, in this case 4800 ppi. Calculate how many pixels
there are in a mm. For 4800 ppi, it is 4800/25.4 ~ 189. Now examine
the bar patterns under high magnification in Photoshop. Set the ruler
scale to pixels. Find the finest bar pattern you can resolve, and count
how many line pairs you see in a certain number of pixels. Suppose you
count 5 line pairs in 30 pixels. Divide the number of line pairs by the
number of pixels and then multiply by the number of pixels in a mm. In
my example, this is (5/30) x 189 ~ 31. You are then resolving about 31
lp/mm. The only ambiguity in this is your decision about which bar
pattern you can actually resolve.


You can also repeat it by scanning at different resolutions and seeing
what numbers you get. The results may vary a bit, but they probably
won't vary dramatically.

Once you have an estimate for lp/mm, you can use that to estimate how
much the scan can be enlarged before the loss of fine detail is obvious
in the resulting image. The usual assumption is that the human eye can
resolve 5 lp/mm at 10-12 inches (but some claim to be able to resolve 8
or more lp/mm). Using that value, for example, a 30 lp/mm image can be
enlarged up to 6 times.

I find it confusing to talk about the scanner really being a 2000 ppi
scanner or whatever. You are trying to specify it as equivalent to
some idealized perfect scanner, but such a scanner doesn't exist. The
problem is confusing scanning resolution, which is the number of samples
you collect, with photographic resolution, which refers to the fine
detail you can actually see in the end result. This is usually measured
in lp/mm as a first approximation, although it would be better to
specify an entire graph of the MTF.

 

[David J. Littleboy]

I think you are making it too complicated. This is the way I've done
it. Scan a test chart with resolution bar patterns at the highest
optical resolution, in this case 4800 ppi. Calculate how many pixels
there are in a mm. For 4800 ppi, it is 4800/25.4 ~ 189. Now examine
the bar patterns under high magnification in Photoshop. Set the ruler
scale to pixels. Find the finest bar pattern you can resolve, and count
how many line pairs you see in a certain number of pixels. Suppose you
count 5 line pairs in 30 pixels. Divide the number of line pairs by the
number of pixels and then multiply by the number of pixels in a mm. In
my example, this is (5/30) x 189 ~ 31. You are then resolving about 31
lp/mm. The only ambiguity in this is your decision about which bar
pattern you can actually resolve.

Finally some solid common sense. Thanks!

David J. Littleboy
Tokyo, Japan

 

[Mike Engles]

I think you are making it too complicated. This is the way I've done
it. Scan a test chart with resolution bar patterns at the highest
optical resolution, in this case 4800 ppi. Calculate how many pixels
there are in a mm. For 4800 ppi, it is 4800/25.4 ~ 189. Now examine
the bar patterns under high magnification in Photoshop. Set the ruler
scale to pixels. Find the finest bar pattern you can resolve, and count
how many line pairs you see in a certain number of pixels. Suppose you
count 5 line pairs in 30 pixels. Divide the number of line pairs by the
number of pixels and then multiply by the number of pixels in a mm. In
my example, this is (5/30) x 189 ~ 31. You are then resolving about 31
lp/mm. The only ambiguity in this is your decision about which bar
pattern you can actually resolve.

You can also repeat it by scanning at different resolutions and seeing
what numbers you get. The results may vary a bit, but they probably
won't vary dramatically.

Once you have an estimate for lp/mm, you can use that to estimate how
much the scan can be enlarged before the loss of fine detail is obvious
in the resulting image. The usual assumption is that the human eye can
resolve 5 lp/mm at 10-12 inches (but some claim to be able to resolve 8
or more lp/mm). Using that value, for example, a 30 lp/mm image can be
enlarged up to 6 times.

I find it confusing to talk about the scanner really being a 2000 ppi
scanner or whatever. You are trying to specify it as equivalent to
some idealized perfect scanner, but such a scanner doesn't exist. The
problem is confusing scanning resolution, which is the number of samples
you collect, with photographic resolution, which refers to the fine
detail you can actually see in the end result. This is usually measured
in lp/mm as a first approximation, although it would be better to
specify an entire graph of the MTF.

Hello

I am not too sure how to calculate the line pairs on my test image.
It is sets of 3 parallel lines with rulings up to 81 in RGB and black.
The best I can clearly discern without sharpening is 48.

Mike Engles

 

[Mike Engles]

Hello

I am not too sure how to calculate the line pairs on my test image.
It is sets of 3 parallel lines with rulings up to 81 in RGB and black.
The best I can clearly discern without sharpening is 48.

Mike Engles

Hello
As a postscript to the above I can give you a link to the image scanned
at 4800 dpi.

http://www.btinternet.com/~mike.engles/mike/Widthcrop.tif

Mike Engles

 

[Leonard Evens]

Hello
As a postscript to the above I can give you a link to the image scanned
at 4800 dpi.

http://www.btinternet.com/~mike.engles/mike/Widthcrop.tif

Mike Engles

Okay. I downloaded this and examined it in my photoeditor (the Gimp
under Linux). Since there are three bars, it is a bit ambiguous, but
counting the space next to the last bar, looking at the pattern marked
48 with the bars vertical, I can count 3 line pairs in a space of
between 15 and 20 pixels. I assume this was scanned at 4800 ppi, which
is the same at 189 pixels per mm.
Using 15 pixels for the spatial dimension yields (3/15) 189 = 37.8 lp/mm
Using 20 pixels for the spatial dimenison yields (3/20) 189 = 28.35
lp/mm.

Doing over again with that sample, I think I can count 2 line pairs in
10 pixels, which would confirm the upper value of close to 38 lp/mm.

Doing the same thing for the one marked 28, I think an estimate of 3
line pairs in 20 pixels is certainly an upper limit. The same
calculations gives for that about 28 lp/mm.

Again doing it over again with that sample, I think I can count 2 line
pairs in 13 pixels, and that gives 2/13 189 ~ 29 lp/mm.

With only at most 3 line pairs to count, this is obvious going to be
subject to significant error. You may count differently. But the
results we seem to be getting are consistent with other patterns I've
downloaded. The best I was ever able to get from the Epson 3200 was
around 31 lp/mm. So the 4870 does appear to give an improvement of 20
percent in one direction, but perhaps not in the other. This is of
course significantly less than the 50 percent increase in scanning
resolution.

Another caveat is that there does seem to be some sample to sample
variation in these scanners. Others have achieved slightly better
results than I have with the Epson 3200.

 

[Mike Engles]

Okay. I downloaded this and examined it in my photoeditor (the Gimp
under Linux). Since there are three bars, it is a bit ambiguous, but
counting the space next to the last bar, looking at the pattern marked
48 with the bars vertical, I can count 3 line pairs in a space of
between 15 and 20 pixels. I assume this was scanned at 4800 ppi, which
is the same at 189 pixels per mm.
Using 15 pixels for the spatial dimension yields (3/15) 189 = 37.8 lp/mm
Using 20 pixels for the spatial dimenison yields (3/20) 189 = 28.35
lp/mm.

Doing over again with that sample, I think I can count 2 line pairs in
10 pixels, which would confirm the upper value of close to 38 lp/mm.

Doing the same thing for the one marked 28, I think an estimate of 3
line pairs in 20 pixels is certainly an upper limit. The same
calculations gives for that about 28 lp/mm.

Again doing it over again with that sample, I think I can count 2 line
pairs in 13 pixels, and that gives 2/13 189 ~ 29 lp/mm.

With only at most 3 line pairs to count, this is obvious going to be
subject to significant error. You may count differently. But the
results we seem to be getting are consistent with other patterns I've
downloaded. The best I was ever able to get from the Epson 3200 was
around 31 lp/mm. So the 4870 does appear to give an improvement of 20
percent in one direction, but perhaps not in the other. This is of
course significantly less than the 50 percent increase in scanning
resolution.

Another caveat is that there does seem to be some sample to sample
variation in these scanners. Others have achieved slightly better
results than I have with the Epson 3200.

Hello

Thanks, I came to similar conclusions.
I cannot decide whether to take up the issue about the variation in
resolution in the X and Y axes, with Epson or not.

I only intended to use the scanner as a general flatbed and to scan my
645 images. I had no great expectations beyond 2400 DPI.

It is after all a consumer scanner and ICE is the best reason for buying
it.

Mike Engles