|Back to main - gPhoto Slimshot Driver - What's Happening|
Added support for Logitech pocket digital into CVS at sourceforge. It's been a while since the last update, due to new job etc. Anyway - Matthias Scherrer provided some very useful USB dumps, so some progress made on the above - it turns out it uses a different packet size, and protocol (ok, it's pretty much... different) from the ultrapocket clones, but not different enough that it merits a seperate driver.
An interesting note is that it returns (AFAICT) the same data for a 1024x768 image as for a 640x480 - I know they say software assisted, but, cmon guys! I mean....
All the interesting (read published ;) algorithms for interpolation seem to be patented. It's a hard world. Been working on one which uses a different approach to anything I've seen published, but of course the results aren't perfect as I have to shy away from some obvious (and therefore patented) stuff. Bah. Americans.
One thing which has been bugging me are some interpolation artifacts I've been seeing, but (in retrospect a bit obvious *sigh*) these seem to be the camera's fault. Green pixels on red and blue lines appear to respond slightly differently, hence I get striping in my images. I suspect this is due to a charge leakage.
Examples, (green pixels only) using Kurt Garloff's demosaicing algorithm (dem), mine (lee), and the original pixels uninterpolated (original). Striping is quite visible
Found a paper which gives several algorithms for better interpolation of Bayer demosaicing than bog standard bilinear. I think a constant-hue goal interpolation produces the best results given the low resolution of the camera.
(what the standard gphoto bayer function uses)
|Gradient based interpolation|
(Still a long way to go, though I've not modified the levels yet)
Found an interesting (though not much use to me! ;) teardown report on the camera.
Lee Benfield - 2003