What are the ethical limits for the application of Photoshop?
I have been intrigued by some of
the image forensic analysis work of Dr. Neal Krawetz as documented in his
blog the Hacker
Factor. He
applies a variety of tools to try to uncover manipulation in digital images and
has several posting dissecting various still photos and videos. One of the
tools he applies is something called the Error Level Analyzer. The original
author of this tool (P. Ringwood) previously had posted the source and binary a
few years back (at which time I downloaded this). A
website now exists
where you can post your images and have them analyzed using this tool.
My personal interest has
been out of curiosity of many images I see posted on Flickr or submitted in
various photo contests that I believed to be doctored in a misleading
manner. Some of these more popular or winning images to me look like some
blatant manipulation has been involved that crosses my personal ethical line
of what should be modified without disclosure. However, trying to
prove such manipulation can be difficult. There is also a debate that
I have with myself and others as to what should be considered ethically
acceptable versus what crosses the line. One of my concerns is that more blatant
modifications will result cheapening the field of photography, and turn it
into more of a field for those with the best graphics artists skills. The
other is the undermining of the hard work by a photographer to capture a
unique image in an honest manner. Consumers and
viewers of such art may become more jaded and question more and more geniune
work from that in which the capture was "cheated" and not necessarily know
the difference.
My interpretation as to what is OK
with the intent to reflect the realism of a scene would consist
of edits that follow these guidelines:
- Fixes for spots such as dust specs.
- Cropping or scaling the image to create a better composition or
adjustment for the target frame size and its aspect ratio.
- Removal of small and distracting, yet superfluous element(s) such as sticks, rocks, or pieces of
litter.
- Exposure adjustments via dodging, burning, curves, exposure masks, etc. for the purpose of better
balancing the overall exposure and maintain sufficient detail in the scene.
- Sharpening to the point that looks natural and attractive.
- Color adjustments to get the white balance correct, to fix a shift in colors due to a change in
the contrast, or adjustment in saturation that maintains a natural look and attractiveness.
- A conservative/modest application of high dynamic range techniques via merging multiple exposures of the same image
(i.e. bracketed exposure sequence). This can be necessary when
photographing a scene like a sunset where the camera simply does not
have the dynamic range to maintain detail in the foreground without
blowing out the highlights. If I apply this technique, I try to maintain the realism of the scene and avoid
those garish colors that some of the canned HDR tools tend to produce.
This often requires additional modifications to the contrast or colors
to restore the scene to how it was believed to look at the time of capture.
- Stitching of multiple images to create a panoramic result. This can also require some work to
warp the image or adjust the level properly. Such distortion in
panoramic captures is typical and can happen as a result of non-level
setup of the camera on the tripod during rotation, or as a result of using wider
angle focal length. Sometimes I bring a leveling base for my tripod with
a bubble level on the camera to avoid the need for as much correction in
the post-processing phase, but sometimes this is not as practical when
traveling lighter.
Some readers of this article may view this as a long
list of liberties. But overall my goal is to maintain the realism of
the scene at the time of the original capture with the edits in line with
adjusting for deficiencies in the technical limitations of the camera and/or
related equipment. Any serious photographer I am aware will find
the above list to be necessary work as part of producing high quality
images. In addition, if you shoot in the camera raw format, some of
this is necessary as part of the conversion to tiff, psd, jpeg, png, etc
anyway.
The difference is the post-processing is being down during the raw
conversion by the user in software manually versus the conversion being done in the camera
by its firmware as
controlled by one of the picture profile settings.
To be clear on my
views of what I think crosses the ethical barrier for non abstract or
impressionist types of images without disclosure, I would treat the
following as crossing the line:
- Introducing
fundamentally new objects into the image for the purpose adding to its
beauty or esthetics. This could be a bird, a moon, a person, or something like that.
For example, I was dumb struck to see a cover photo of the July 2009 issue of
Outdoor Photographer with a full moon placed in the middle of
Delicate Arch. Having been to this location on multiple occasions I was
aware of the angle in which it was taken. Unless there was some
shift in the orbit of the moon, such an image would not be possible.
In addition, the scale and lighting on the moon was obviously out of
whack (the moon could not be this size from the distance the
photographer was to the arch). They may as well have cloned in a rainbow or unicorn
while they were at it :^)
- Adding a different sky to the
background that did not originally exist. Some may tweak the clouds or
clone in a few to potentially fix an over-exposure problem (i.e. fringe
edge of an existing cloud is blown out in the highlights), and that may be more reasonable. But to create a
sky from a completely different image crosses the line.
To aggregavate matters, I do find it a bit frustrating that some requests for photo submissions make
blanket statements such as: "No photoshop", without in my view an understanding
of the distinction between these two different class of edits described
above. Perhaps
it would be hard to judge or referee such a distinction, so as a result a
zero tolerance policy is stated instead?
Anyway,
now that I have gotten that off my chest, I
thought I would perform an experiment on one of my one photographs to see if
I could use the Error level analyzer tool previous referenced to determine if doctoring
of the image could be
detected. As shown to the right, I started with an image that I
captured on one of my bird photography outings. I liked the lighting on the
birds as it was difficult to get this opportunity with them flying somewhat
toward me and into a sunrise. But my
framing at the time of capture was leading a little to much to the right.
Shooting birds in flight can be tricky while still maintaining focus lock as
you only have seconds to get it right. Unfortunately
this resulted in a slight clipping of the end of one of the Sandhill Cranes legs and
wingtips as shown to the right.
Rather
than throw the photo out, I liked the image enough that I applied some
photoshop work to restore the clipped portion and extend some space to the
left and above
to better balance the composition. You can see the results of my
handy work here in this next version.
This required a bit of
finesse involving
cloning a foot and some wing tips from one of the other birds and applying
it to the left-most bird. This should also give the reader an idea of
the types of things that are easily possible for those skilled in the art.
To me, this level of manipulation I have shown here is a gray area. I believe it would be
OK to reference such a result for documentary purposes (i.e. for a book or
magazine documenting the migration of Sandhill Cranes). But I don't believe
it would be proper to submit such an image to a photo contest or some other
judged artistic competition without at least disclosing such edits, since the
fundamental skill in winning the contest should be based on your
photographic skills, and not your Photoshop skills. If it was a
Photoshop contest, then the circumstances would be different.
After attempting to analyze the results using the error analyzer tool, it
was not that obvious that the edits could be detected on the bird itself.
The background area to the left and top that I expanded by cloning the
immediately adjacent area is an obvious give away using this tool.
Original unedited ELA result:
Edited (doctored) ELA result:
According to the ELA website, this
tool is supposed to work by introducing a small amount of noise and then
highlighting the pixel areas that that already have reached some error
minima. Theortically false introduced content will have less error on the
borders where the changes have been made. But based on this algorithm it is
not clear how effective it is at detecting this type of manipulation.
Based on the edited ELA result, I do notice the rear foot does have a small
amount of increased illumination, but then again, I knew to look there. It
is interesting to note the feet of the right-most bird also exhibits some of
this increased illumination, although I did not edit that area of the image
(honest :^). Perhaps one weakness with a tool like ELA is that its
algorithms may be better at catching manipulated JPEG compressed image
content versus those maintained in raw. A raw file will have more bits
of precision per pixel, and thus less likely to saturate its numerical
range.
So at this point, I have not reached any conclusions on a more robust means
for detecting such edits through a tool like ELA. As an example, there is a particular image I note on
Flickr titled "Capocotta
Beach" that I am curious about. The artistic aspects of this image
appeals to me, but the seagull looks suspicious. The lighting on it seems
unnatural for what I would expect to exhibit more backlighting
characteristics (Apologies to the author if I am wrong.) Using the ELA
website to examine it does show a little unusual highlighting around the
beak, but it hard for my untrained eye to know the difference of this from
similar randomness in the ELA output on the sand of the beach. I have not
posted it out of respect for the owners copyright, but you can go through
the exercise yourself by viewing the image at the large size in source form
in your browser, and then cut-and-paste the jpg reference as a link to the
ELA website to perform the analysis.
