Warning: get_object_vars() expects parameter 1 to be object, null given in /home/newspeak/public_html/wp-includes/taxonomy.php on line 324
February 09
So, I work in a research lab, so when I got a smudge on my camera lens this morning, I just grabbed a washbottle of isopropanol and squirted in on, wiped it off with lens paper, and blew it off with dusting gas. It probably says explicitly in the manual not to use solvents on the lens, but I figured “fuck that, I’m a scientist, I know what I’m doing!”
It turns out it was a bad idea:
(warning: nerd-out inside)
I was shooting at 8 MP, and I can’t upload a file of that size here, but the effect is even more dramatic when the image is downsampled to 800×600 with no interpolation. The image is otherwise unprocessed. At full resolution, there are just some jaggies where it seems to have shifted each scanline by a few pixels to the left or right, and darkened every fourth one or so (on this more below). The artifacts you see here are essentially a Moiré pattern from that pixel-jiggling with the pixels (approximately every third one) it picked for the downsampling. Next, a detail of the desk area at full resoltion: 
Talking with a friend about this, the best guess is that the alcohol got the sensor wet and was shorting out the CCD in some weird manner. I don’t really know how those things work, but there’s gotta be some crazy serial/parallel bus to truck the samples off the sensor, and somehow the liquid munged this up in a weirdly periodic fashion. For him and the one other person who might be reading who I know will get this, I took an FFT of the (full-resolution) image. I downsampled this as well (to 800×600, actually I think downsampling an FFT at higher resolution results in a second-level discrete wavelet transform…. right?), resulting in the following: 
Definitely some abnormal spectral components in there, and like I said, I have no real grasp of camera electronics or how I got such weird periodic artifacts. By this point I was taking pictures randomly, just checking out the ensuing weirdness. The theory that this messed up the sensor is corroborated by one of the next pics: 
The first picture above is way over-exposed, which it shouldn’t be, because I was shooting on maximum-dumbshit mode. By this point, though, the camera was developing some serious exposure problems (again, except for resizing, this is straight off the camera). Also, I was firing off exposures as quickly as I could, since none of this crazy behavior was showing up on the preview screen.
Immediately after this picture was taken, everything started returning to normal. As far as I can tell now, it’s like nothing ever happened. I’m not sure whether to be happy that I didn’t break my camera or sad that the weirdness went away. I invite you all to get a cheap camera and fuck with it and see if you can come up with anything bizarely similar to this. Cameratoss and HDR are really cool tricks, but they’re fundamentally analog. Nothing like the morning that platnoic form shared by DEVO and kid606 put on a private concert in my CCD.
Posted by: Andrew Hutchison in FoToes, Science | Permalink
Comments
4 Responses to “Broken camera fun!”
Leave a Reply
-
-
-
Warning: get_object_vars() expects parameter 1 to be object, null given in /home/newspeak/public_html/wp-includes/taxonomy.php on line 324
You warned me, so it’s my own fault, but wow. That was really nerdy.
On a totally unrelated note, Aaron you definitely absolutely should not read this before heading up north:
http://www.somethingawful.com/d/news/wood-cutting-mishap.php?page=1
How does the FFT image translate to the regular image? Frequency of what vs. what on what scale? What does a pixel on the FFT image represent exactly?
NP, good question. The units for the axes of the transformed data will simply be 1 over your original axis units. For example, in this case you are plotting the Fourier spectrum against inverse pixels (1/pixels). That’s the “frequency” scale. A spike in the spectrum means you have a strong modulation at a certain pixel periodicity.