186 lines
7.7 KiB
HTML
Executable File
186 lines
7.7 KiB
HTML
Executable File
<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
|
|
<html>
|
|
<head>
|
|
<title>The i1pro Hi Res. Mode</title>
|
|
<meta http-equiv="content-type" content="text/html;
|
|
charset=windows-1252">
|
|
<meta content="Graeme Gill" name="author">
|
|
</head>
|
|
<body>
|
|
<h2 style="text-decoration: underline; font-weight: bold;">Does the
|
|
i1pro High Resolution mode improve accuracy ?<br>
|
|
</h2>
|
|
A question that has been asked is : "<span style="font-weight:
|
|
bold;">You've extended the Eye-One Pro with a high resolution
|
|
spectral mode, giving readings at 3.3nm spacing rather than the
|
|
default 10nm. Does this mode improve accuracy ?</span>"<br>
|
|
<br>
|
|
This is a quite reasonable question. The following attempts to
|
|
answer it.<br>
|
|
<h4 style="text-decoration: underline;">Why would a higher
|
|
resolution spectral mode improve accuracy ?<br>
|
|
</h4>
|
|
A spectrometer computes CIE tri-stimulus values by measuring
|
|
spectral values and then weighing those values by the observer
|
|
curves before summing the weigted values. The accuracy depends on
|
|
the correct weighting being applied at each wavelength. If the color
|
|
is composed of very narrow spectra peaks, as is sometimes the case
|
|
for certain light sources and many display devices, then the exact
|
|
positioning of one of the peaks on the observer curves may be
|
|
influencial in the final color value, and too coarse a quanization
|
|
of the spectral readings may lead to tri-stimulus errors. So in
|
|
theory increasing the spectral reading resolution to 3.3 nm should
|
|
lead to improved color accuracy with narrow spectra color sources. <br>
|
|
<h4 style="text-decoration: underline;">Why may this not work in
|
|
practice ?</h4>
|
|
<p>The instrument spectral resolving power is set by a number of
|
|
factors, and a critical one is the entrance slit width. By
|
|
measuring a very narrow band source such a as a laser, using the
|
|
default 10nm resolution indicates a FWHM (<a
|
|
href="http://en.wikipedia.org/wiki/Full_width_at_half_maximum">Full
|
|
|
|
|
|
|
|
|
|
|
|
|
|
width at half maximum</a>) of about 25nm. Doing a measurement at
|
|
3.3nm resolution reveals that the optical limit seems to be about
|
|
15nm, so there is some hope of improvement from that perspective.</p>
|
|
<p>Another factor is that the calibration data for the instrument is
|
|
only given at 10nm intervals. So to produce calibrated readings at
|
|
3.3nm intervals, it is necessary to up-sample the calibration data
|
|
with sufficient accuracy. If the calibration data is sufficiently
|
|
smooth (indicating that the underlying device characteristics are
|
|
also smooth), or any slight inaccuracy will get calibrated out
|
|
(which is typically the case for reflective measurements) then
|
|
this may not be a limitation either. In the case of the i1pro2,
|
|
which seems to have a diffraction grating/light sensor with a less
|
|
smooth spectral efficiency curve than the Rev A - D models, the
|
|
task of up-sampling the emissive calibration data with sufficient
|
|
accuracy is more difficult.<br>
|
|
</p>
|
|
<h4 style="text-decoration: underline;">The verification experiment<br>
|
|
</h4>
|
|
To give some indication of whether ArgyllCMS's high resolution
|
|
spectral mode is capable of improving color measurement accuracy, or
|
|
at least to indicate that it doesn't noticeably worsen it, the
|
|
following fairly simple, real world experiment was performed:<br>
|
|
<br>
|
|
A measurement target consisting of white + primary + secondary
|
|
colors (White, Red, Green, Blue, Cyan, Magenta, Yellow) repeated 10
|
|
times was used. This target was displayed on a conventional LCD
|
|
screen with a CCFL backlight (MacBook display), and measured using
|
|
using ArgyllCMS V1.6.0 <a href="dispread.html">dispread</a>:<br>
|
|
<br>
|
|
1) Using a <a
|
|
href="http://www.jeti.com/cms/index.php/instruments-55/radiometer/specbos-1211">JETI
|
|
|
|
|
|
|
|
|
|
|
|
|
|
specbos 1211</a> reference Tele-Spectro-Radiometer.<br>
|
|
<br>
|
|
2) Using an i1pro2 in standard 10nm mode.<br>
|
|
<br>
|
|
3) Using an i1pro2 in ArgyllCMS 3.3nm mode.<br>
|
|
<br>
|
|
The resulting readings were then analyzed using <a
|
|
href="colverify.html">colverify</a>.<br>
|
|
<br>
|
|
The results were analyzed two ways, first in absolute value error
|
|
terms, and secondly in brightness (Y) normalized terms, the latter
|
|
corresponding to the typical way such readings are used for display
|
|
calibration and profiling. <br>
|
|
<br>
|
|
A second, similar experiment was run on a CRT type display.<br>
|
|
<h4 style="text-decoration: underline;">Results:</h4>
|
|
<p><br>
|
|
LCD display:<br>
|
|
</p>
|
|
<p>Absolute errors of i1pro2 10nm mode to specbos 1211:<br>
|
|
</p>
|
|
Total errors (CIEDE2000): peak =
|
|
3.070420, avg = 2.204137<br>
|
|
<br>
|
|
Absolute errors of i1pro2 3.3nm mode to specbos 1211:<br>
|
|
<br>
|
|
Total errors (CIEDE2000): peak =
|
|
2.108411, avg = 1.568577<br>
|
|
<br>
|
|
<br>
|
|
White Y normalised errors of i1pro2 10nm mode to specbos 1211:<br>
|
|
<br>
|
|
Total errors (CIEDE2000): peak =
|
|
2.419800, avg = 0.747926<br>
|
|
<br>
|
|
White Y normalised errors of i1pro2 3.3nm mode to specbos 1211:<br>
|
|
<br>
|
|
Total errors (CIEDE2000): peak =
|
|
1.595033, avg = 0.578270<br>
|
|
<br>
|
|
<br>
|
|
So in this particular situation, hi-res mode improves accuracy by
|
|
somewhere between 0.2 and 0.6 DeltaE 2K.<br>
|
|
<br>
|
|
<br>
|
|
Example of white spectrum for the three measurements (red: 10nm
|
|
i1pro2, green: 3.3nm i1pro2, black: specbos):<br>
|
|
<img alt="specbos 1211 (Black), i1pro2 10nm (Red), i1pro2 3.3nm
|
|
(Green)" src="i1proHiRes.jpg" height="335" width="667"><br>
|
|
<br>
|
|
<p><br>
|
|
CRT display:<br>
|
|
</p>
|
|
<p>Absolute errors of i1pro2 10nm mode to specbos 1211:<br>
|
|
</p>
|
|
Total errors (CIEDE2000): peak =
|
|
1.516886, avg = 0.965740<br>
|
|
<br>
|
|
Absolute errors of i1pro2 3.3nm mode to specbos 1211:<br>
|
|
<br>
|
|
Total errors (CIEDE2000): peak =
|
|
1.751776, avg = 0.887878<br>
|
|
<br>
|
|
<br>
|
|
White Y normalised errors of i1pro2 10nm mode to specbos 1211:<br>
|
|
<br>
|
|
Total errors (CIEDE2000): peak =
|
|
1.509129, avg = 0.654752<br>
|
|
<br>
|
|
White Y normalised errors of i1pro2 3.3nm mode to specbos 1211:<br>
|
|
<br>
|
|
Total errors (CIEDE2000): peak =
|
|
1.284044, avg = 0.622501<br>
|
|
<br>
|
|
<h4 style="text-decoration: underline;">Conclusions:</h4>
|
|
The results for the conditions of this particular experiment
|
|
indicate that ArgyllCMS High Resolution mode can very slightly
|
|
improve colorimetric measurement accuracy of display devices.
|
|
Accuracy may conceivably be improved a little more than indicated by
|
|
this experiment for i1pro rev A-D instruments which have a smoother
|
|
diffraction grating/light sensor characteristic, but it is also
|
|
conceivable that an unfortunate combination of display spectra and
|
|
the i1pro2 may result in reduced accuracy. More extensive testing of
|
|
a range of instruments and illuminants would be needed to allay such
|
|
a concern. <br>
|
|
<br>
|
|
<b>In summary</b>: the High Resolution mode is unquestionably useful
|
|
for showing more spectral detail, and demonstrates promise of
|
|
improved accuracy, but should probably not be used used for
|
|
colorimetric measurement when the highest possible confidence is
|
|
desired.<br>
|
|
<h4 style="text-decoration: underline;">Raw Data:</h4>
|
|
The raw measurement data is available in this <a
|
|
href="i1proHiRes.zip">.ti3 archive</a>.<br>
|
|
<br>
|
|
<br>
|
|
<br>
|
|
<br>
|
|
<br>
|
|
<br>
|
|
</body>
|
|
</html>
|