On 16 July 2008 I did some experimentation to compare various methods of
exposure compensation for my Olympus E-510 camera. I took five photos of the same scene,
with exposures bias of -2, -1, 0, 1 and 2 EVs, with both raw and JPEG images, and then tried
to correct the exposure with four methods. I can't display a raw image directly on the web,
so the following groups of 7 images are the results. Exceptionally, some of the originals
are reduced by 75% linearly: the detail isn't important.
The first photo is the JPEG image as loaded from the camera.
The next one is the result of my attempts to fix things up manually with
xv and the JPEG image.
Then comes one done with xv and the histeq function.
After that, an image created from the raw image
with ufraw by adjusting the exposure
value back by the same amount I had offset in the first place.
The results of passing the raw image through ufraw-batch (really the same
program as ufraw) with the parameter --exposure=auto, which should do the
same thing as the preceding one.
On 15 June 2012 I revisited this page and did two further conversions
of the raw images with DxO Optics "Pro", one with the default parameters and one with the “HDR Artistic”
“preset”, which I have found gives good results with this kind of image.
Results by exposure value
Image underexposed by 2 EV
If you have JavaScript enabled, running the cursor over images in this section will give a
direct comparison with the original (first) image.
The obvious problem here is that the contrast is strongly dependent on the exposure. I can't
find a way to do any better with xv. In particular, overexposed images look hardly any
better, maybe even worse.
The biggest problem here is the change in colour saturation. The underexposed images come out
better, but the lack of detail is apparent in the shade. There's probably nothing to be done
here, since the detail was probably never there.
These are the ones done simply by changing the exposure back by the same amount by which it had
been misexposed. The results look quite good for the underexposed images, and surprisingly
washed-out for the +2 EV overexposed image.
This even manages to make the original image worse! It does probably the best for overexposed
images, but worse than the manual adjustment for the underexposed images.
These produce what I consider the best results for this kind of subject. In particular, the
shadow detail is better than any of the others. Not surprisingly, it handles the underexposed
images best, though they're clearly still inferior to correctly exposed images.
It's clear that using the raw image gives better results, but they're not spectacular. In the
darkroom, 2 EV is not a big problem to compensate for. With modern digital sensors, it seems
to be so. Like with diapositive film, it's also clear that it's better to underexpose than to
overexpose. And the best way to do things is probably still manually with ufraw; there
are less mechanical approaches than by EV.
Comparing the processing tools, it's clear that DxO wins. While the default results barely
differ from the in-camera JPEGs, the
“artistic” profile really lifts the shadows and makes even the -2EV image look almost
acceptable.
After that, it's not so clear. With xv I was able to get better results from the
underexposed images, at the expense of some burnout in the highlights. ufraw's
auto-exposure compensation appears useless, and the manual use of ufraw is not very
convincing. In particular, it handles the underexposed images badly.
It's also interesting to note that the raw files aren't much larger than the high-quality
JPEGs, so I can move to using only raw images:
Despite claims on the web, these raw
files must be compressed; otherwise they would be over 20 MB in size. That should save me
having to buy a larger flash card for the camera.
It occurred to me that the tests above left out the obvious contender: the “Master
2” software provided by Olympus specifically for the camera. Somehow I have prejudices
about that sort of thing, but today I decided to try it out anyway, on boskoop, my old
G4 Apple. It did nothing to dispel my prejudices.
Admittedly, boskoop is not exactly state-of-the-art, but even so, taking 20 seconds
of CPU time to start up is a bit extreme. Then it continued with an animated help screen,
showing nothing more interesting than an adapter card being pushed into the side of a
laptop, along with a text in a too-small window with a scroll bar, using about 90% CPU
time—even when iconified! And that on a desktop machine to which I connect the camera
with a USB cable:
This help brings with it an obligatory way of looking at the world that is completely
different from mine. “How to Enjoy Slide show?”, for example. But where are my
files? “How to Browse Images/Movies?” looks like the right one to choose, but
it just tells you (in four screens with one or two sentences each) how to work your way
around the “Browse” window. The only way I could find to actually get new
images is from the camera. It seems that these people just don't understand file systems.
Spent some time looking for a way to look at existing images; it doesn't seem to cater for
that.
Took a look at the “Browse” screen, which offered a function
“transfer”, which does exactly that, even when not necessary: it seems that to
process photos, it first needs to copy them to an “Album”, and that it
did—all 650 MB of the photos I took on Sunday—once I found a way to type in the
name: it wanted me to select from a graphical representation of “folders” with duplicate names:
The directory I'm looking for was NFS mounted as /dereel/home/grog/Photos/20080713,
so it would be reasonable to assume that dereel in that second image would be the
place to look. But no, that's a different file system. This is one of the two entries
marked home. Which one? No idea.
Fortunately, if you close that window you get the opportunity to enter the directory name
manually, without any help like file name completion, and without the option of globbing for
specific files only (at least, I couldn't find one). After only 15 minutes and the best
part of a gigabyte of data, I was in the position that I am by default with real software.
Well, almost. After a bit of searching (WHY do you always have to search for
these things? WHY can't they just tell you the path?), discovered:
=== grog@boskoop (/dev/ttyp1) ~/Pictures 8 -> l -rt
total 193
...
drwxr-xr-x 3 grog 1000 102 Jul 18 18:40 OLYMPUS Master 2
=== grog@boskoop (/dev/ttyp1) ~/Pictures 9 -> cd OLYMPUS\ Master\ 2/
bash: cd: OLYMPUS: No such file or directory
These horrible path names with spaces in them, which continually mess up scripts. In
my case, I define cd as a bash function, and it trips over spaces even if, as
in this case, they're quoted.
And of course the files were transferred in alphabetical order without maintaining the
modification time stamps (even though they all had EXIF data), so I ended up with them all
out of sequence.
Apart from that, the software seems to offer much the same functionality as ufraw.
It has some things, like compensation for chromatic aberration, that I don't
think ufraw has. But who can bear to use that sort of thing?
