January 2006
Over recent weeks it has become apparent that there has
been a very significant increase in the number of visitors reading this
page. I'm at a loss to explain this increase in interest, especially
since the subject of the tutorial and the cameras to which it pertains
has long been superseded. Clearly, interest in the topic still exists,
but it would be remiss of me not to point out a few important changes
that have occurred over the period since the tutorial was first
published.
Both cameras and software have came and gone, likewise
the need to even bother with the machinations described. Why should we
not be bothered anymore? The answer is simple - Adobe Camera Raw has
facilitated "highlight recover" since version 2.2. This feature actually
uses the subtle colour imbalances in the highlights to create extended
neutral highlights. Similar features are now available from other vendors. Nevertheless, I
believe Adobe Camera Raw is the superior solution, especially version 3.0
and higher, which includes even more advanced highlight recovery
algorithms.
If you're interested in learning more about Adobe Camera
Raw then I recommend that you read Real World Camera Raw by Bruce Fraser.
This book, now in it's second edition, is without doubt the best
available guide to raw image editing. Real World Camera Raw is available
from Amazon.com and other major book sellers.
Notice the spread of image data from the shadow end (left
hand side) to the highlight end (right hand side). This actual image when
converted to TIFF or JPG using Canons File Viewer Utility (FVU) or Raw Image
Converter (RIC) application results in areas of the of the sky having
a mild magenta colour cast. This colour cast cannot be removed without
significant colour shifts in other parts of the image. Even exercising a
great deal of care when creating an ICC/ColorSync profile to control the
highlights found within such images can be extremely difficult. I don't
know for sure but my guess is that the nature and/or location of the
spike
(value on the Histogram levels scale) and run-on data might have some influence on the final
result. The spike is well short of level 255, but it along with the
run-on data suggests that the sensor has gone into saturation. I've found
that when this spike appears in other linear raw images then the highlight region of the converted
images have a colour shift. Other Canon D30 users have reported similar
experiences.
The following procedure (really a kludge) is one
of a few that I have adopted when profiling
the Canon EOS D30 in linear raw mode. Although, I don't need to employ it
for all my camera profiles.
In this tutorial I assume the reader is using profiling
software that has been designed for the purpose and is capable of reading
the GretagMacbeth ColorChecker or ColorChecker DC charts.
The software used when writing this tutorial was
Pictographics inCamera
Professional 3 and is available for both Mac and PC platforms.
Nevertheless, the method discussed is also perfectly feasible for use
with other profiling software. I've tried it with ProfileCity ICC-DCam
and Praxisoft CompassProfile Camera.
Figure 4 - Histogram of Figure 3
If we compare the histograms Figure 4 with
Figure 5 (Histogram of the original image in Figure 1) we can
immediately see that there is a significant shortfall of image data for
the ColorChecker DC Chart capture. Our first reaction would be
that the DC chart has been under-exposed. However, the profiling target is
correctly exposed (if the image was fully processed by Canons own
applications the max white patch would have a value of 242). So why such
a major difference in the two histograms; what does it mean for our
profile, and how do we get out of the potential mess this can cause? We
could write a book and not answer these questions - Answering the
first is way beyond this tutorial and so I'll concentrate on 2 and 3,
which are really the only ones that matter.
Figure 5 - Annotated Histogram
of Figure 1
Our questions:
Q. 2 - What does the discrepancy in data spread mean for our profile?
Ans. 2 - It should be pretty obvious that the
profiling target isn't representative of the real world. This means that
the profiling software has NO means of determining how the camera
sensor performs in the highlight region and so in all likelihood it will
produce a camera profile that causes blown highlights and/or exaggerate colour imbalances in the highlight region.
Q. 3 - How do we get out of the potential mess this
discrepancy in data spread can cause?
Ans. 3 - In the example to follow I will use a
compensation curve (sometimes called a tone reproduction curve TRC).
There are other methods but this is the one that I've found to produce the
most consistent results.
The compensation curve is really the key to
getting our profile correct. We need to apply a compensation curve to the
captured test chart so that the data spreads more fully over the range 0
to 255. However, in order that we don't trash the image at the outset it
is ESSENTIAL that the compensation curve and ALL
subsequent editing, profile assignment and conversions are carried out in
16bit mode.
The true value for the maximum white patch of the
ColorChecker and ColorChecker DC charts is Lab=96
(approximately level 242 when working in a gamma 2.2 working space); so we want a compensation curve that achieves
this whilst maintaining the correct relationship with other patches on
the target chart. A typical compensation curve for the D30/60 is shown
below, but I've read of other folk who have used 2 much shallower curves
applied in succession. It's also worth mentioning that a simple gamma 2.2
curve appears useless for this exercise - don't ask me why. However, to
my mind, it
implies that the highlight response of the D30/60 sensor isn't as linear as Canon make out.
Getting a compensation curve that satisfies your specific
requirements requires some experimentation and remember you also need to
save it for use with ALL subsequent images that you intend using
the linear raw profile with.
Figure 7 - Typical Compensation
Curve
The result of applying the above compensation curve to my
capture of the ColorChecker DC target is shown below and from the
associated histogram it should be obvious that we are now looking at
something that more closely resembles the spread of data for our original
image as shown in Figure 2.
Figure 7 - ColorChecker DC Chart
with Compensation Curve
Figure 8 - Histogram of Figure 7
So how does the above curve impact upon my real world
image? Hopefully the screenshots below should give you some idea.
Figure 9 - Original with
Compensation Curve
Figure 10 - Histogram of Figure
9
Notice that the same compensation curve had a much less
dramatic influence on the spread of data in the real world image than it
did on the profiling target. The highlight end has almost reached level
255 but almost is a long way from doing so in terms of our final image -
we won't have blown highlights. Using this technique I've also found that
the colour imbalances caused by apparent sensor non-linearity are eliminated in
the vast majority of cases.
Now that we have our target conditioned for use in the
camera profiling software it is simply a matter of following the relevant
instructions provided by the software vendor. If you don't yet have such
software I offer some guidance as part of my review of
Pictographics inCamera
Professional.
Once the camera profile has been created it necessary to test
it. Typically the method discussed above will result in a profile that
will produce fairly neutral albeit slightly flat images, as shown below.
However, it should be obvious that the camera profile has significantly
improved upon the compensated image shown in Figure 9. If I had a
criticism it would be the image lacks contrast and saturation, but both
are very
easily fixed in Photoshop.
Figure 11 - Original with
Compensation Curve and ICC Profile
One of the big benefits of using inCamera Professional
Pro for this method is the ability to edit the profile and thus
increase the contrast and if absolutely necessary remove any remaining
non-linearity (colour casts) in the highlights.
Hopefully Canon will some day see sense and allow D30 and
D60 users access to the gamma correction curve already built into their
EOS 1D software or recode the D30/60 Firmware and CRW image format so
that user defined Tone Curves can be embedded into the raw images.
