Sudden resolution improvementfor my Epson 3200?

  • Thread starter Thread starter Leonard Evens
  • Start date Start date
L

Leonard Evens

I have been consistently getting at best 28-30 lp/mm from my Epson 3200.
I just scanned some 4 x 5 HP5+ b/w film. Two frames have windows
with blinds, I can count 7-9 blinds in a space of 20 pixels. If my
arithmetic is right, this comes out to close to 40 "blind pairs" per mm.
Of course blinds don't constitute a line bar pattern, but they do
consist of pairs of white blinds and a black spaces next to them. The
black space is narrower than the white blind. Perhaps such a periodic
pattern is easier to resolve than a standard bar pattern, but on the
other hand, the contrast is fairly low. The film was somewhat
overdeveloped relative to normal conventional printing to take advantage
of the scanners dynamic range. The maximum density seems to be about 2.4.

Any comments on what is going on?
 
Leonard Evens said:
I have been consistently getting at best 28-30 lp/mm from my Epson
3200. I just scanned some 4 x 5 HP5+ b/w film. Two frames have
windows with blinds, I can count 7-9 blinds in a space of 20 pixels. If
my arithmetic is right, this comes out to close to 40 "blind pairs" per
mm. Of course blinds don't constitute a line bar pattern, but they do
consist of pairs of white blinds and a black spaces next to them. The
black space is narrower than the white blind. Perhaps such a periodic
pattern is easier to resolve than a standard bar pattern, but on the
other hand, the contrast is fairly low. The film was somewhat
overdeveloped relative to normal conventional printing to take
advantage of the scanners dynamic range. The maximum density seems to
be about 2.4.

Any comments on what is going on?
Click to expand...
What angle are the blinds at? A 45deg test pattern can resolve *up to*
1.4x as much detail as a horizontal or vertical. This will not occur if
the lens system limits resolution in a circularly symmetric manner, but
can if the CCD sensors or sampling are the critical resolution limits.
This 'feature' is one of those that Fuji exploit with their Super-CCD
digital cameras, to get twice the effective number of pixels as the CCD
physically contains.
 
Kennedy said:
What angle are the blinds at? A 45deg test pattern can resolve *up to*
1.4x as much detail as a horizontal or vertical. This will not occur if
the lens system limits resolution in a circularly symmetric manner, but
can if the CCD sensors or sampling are the critical resolution limits.
This 'feature' is one of those that Fuji exploit with their Super-CCD
digital cameras, to get twice the effective number of pixels as the CCD
physically contains.
Click to expand...

They are at a slight angle, but, eyballing it, I didn't think it was
enough to account for the full improvement. But I will go back and
check it out quantitatively.
 
SNIP
They are at a slight angle, but, eyballing it, I didn't think it was
enough to account for the full improvement. But I will go back and
check it out quantitatively.
Click to expand...

If you scan an approx. 5 degree slanted edge you can figure out the edge
spread function, which will allow to calculate the MTF. You can either try
and manufacture such an edge (e.g. slanted razorblade in a slide mount), or
photograph an edge which will give you the MTF of the imaging chain
(including camera shake!). In both cases, try to avoid clipping as it
reduces accuracy.

Alternatively, you can accurately determine the absolute limiting resolution
of your imaging chain by photographing the sinusoidal radial grating
("star") target I proposed earlier. That target has several interesting
properties, including the non-critical shooting distance, multiple
resolution angles and simple quantification.

Bart
 
Bart said:
SNIP



If you scan an approx. 5 degree slanted edge you can figure out the edge
spread function, which will allow to calculate the MTF. You can either try
and manufacture such an edge (e.g. slanted razorblade in a slide mount), or
photograph an edge which will give you the MTF of the imaging chain
(including camera shake!). In both cases, try to avoid clipping as it
reduces accuracy.
Click to expand...

I'm sure I've seen explanations of how to do this before, but it would
save me some time if you would give me a reference explaining how to do
it. I do know the basic mathematics behind it.
 
Leonard Evens said:
I'm sure I've seen explanations of how to do this before, but it would
save me some time if you would give me a reference explaining how to do
it. I do know the basic mathematics behind it.
Click to expand...
There is an explanation and links to source code to implement this at
http://www.normankoren.com/Imatest/sharpness.html

The edge transfer function (ESF, the result of imaging the sharp edge)
is a close approximation to the integral of the line transfer function
(LSF, the response the imaging sensor has to an infinitely thin line of
known intensity. So, differentiate the ESF with respect to distance and
you get the LSF. If you then fourier transform the LSF, the absolute
magnitude of the transform gives the modulation transfer function (MTF)
as a function of spatial frequency.

If the edge is perfectly aligned to the sampling axis then you are
limited to estimating the MTF bt the Nyquist frequency resulting from
the sampling density. In fact, phase effects mean that there is quite a
lot of variation in the resulting MTF in the region of the Nyquist
limit.

However a cunning trick, suggested many years by a colleague of mine who
now works in the microwave imaging field, and critical to the technique,
is to slant the edge slightly relative to the sampling grid. Along the
data axis across the edge this makes little difference, and you are
still essentially limited by the Nyquist. However, in the orthogonal
axis, that closest to running along the edge itself, many more samples
occur as the edge crosses a single sample pitch. The effect is to scale
up the sampling density, and thus the spatial frequency of the resulting
MTF, relative to the edge by the tangent of the tilt angle, and thus
permit you to accurately calculate the MTF up to and many times greater
than the Nyquist limit of the sampling density. Thus, with a 5deg tilt
angle the frequency of the measurement scales up by a factor of 11,
permitting the MTF to be determined up to around 5.5x the Nyquist limit,
by which point the MTF of even the best resolution system will have
fallen to negligible levels.

However, it is important to remember that using a near vertical edge the
*vertical* axis data is to used determine the oversampled *horizontal*
ESF, LSF and subsequent *horizontal* MTF, whilst for a near horizontal
edge the horizontal axis data is used to determine the vertical MTF.
 
Leonard Evens said:
Bart van der Wolf wrote: SNIP

I'm sure I've seen explanations of how to do this before, but it would
save me some time if you would give me a reference explaining how to do
it. I do know the basic mathematics behind it.
Click to expand...

I see Kennedy already explained.
To recap: The Edge transfer is the Edge Spread Function.
Differentiating the ESF will give the Line Spread Function.
The absolute magnitude of the Fourier transform of the LSF is the MTF.

Normally one would differentiate the ESF along the orthogonal axis almost
perpendicular to the slanted edge, and then average many center aligned LSFs
because single pixel wide edge transfers fluctuate due to gradual phase
shifts. The edge transfer can be estimated more accurately by increasing the
sampling density if one differentiates along the orthogonal axis almost
parallel to the slanted edge. The tangent of the slant angle is the measure
for increased sampling density. You'll understand when you see it, if you
don't already.

Bart
 
Back
Top