Tuesday, July 23, 2013
Dual-ISO Mode in Canon DSLRs
Bitbucket's Magic Lantern project comes up with an interesting firmware hack on Canon DSLRs. Apparently, Canon hardware allows changing the readout or ADC gain at a fast pace, every two rows. So, every two rows one can flip the effective ISO, for example, between 100 and 1600 to get an extended dynamic range at the expense of reduced vertical resolution:
Subscribe to:
Post Comments (Atom)
It's not quite as you describe, from what I understand.
ReplyDeleteFrom my reading of the documentation surrounding this hack, there are actually two PGA configuration channels in the analog front end, which read out pixel lines interlaced.
Under normal operating conditions, both channels are set to the same gain. This modification simply sets them differently, which results in alternating scan lines being fed through different gains in the front-end.
Critically, the ISO settings are fixed across the entire frame, it's just that the pixel rows (actually every two rows) are fed through alternating PGAs.
Exactly.
DeleteMixing both fields is proving not straightforward: ISO 100 and 1600 (for example) have very different noise and color characteristics, and the combined HDR image displays some interlaced artifacts (obviously in the highlights and shadows, but also, to a lesser extent, in the midtones). I'm sure a1ex (the ML developer behind this) would appreciate some help with the combination algorithm.
Is this similar to Sony's SVE (spatially varying exposure) approach to HDR?
ReplyDeletehttp://image-sensors-world.blogspot.tw/2013/01/sony-explains-hdr-video-mode-in-its.html
Not exactly. Sony alternates exposure between the rows. In this case the gain is alternating.
DeleteAnd herein lies the rub. By the time they alternate the gain, all that is being affected is the AFE noise (gain stage and the ADC). That's worth something, but not all that much more than just applying that same alternating gain to the output image, digitally. Unlike with the alternate line exposure technique, this method leaves the readout SNR out of the array the same (signal is not increased by extended exposure). It would be interesting to see what SNR numbers they can get out of this method.
DeleteCanon DSLR cameras, unlike those from other vendors, use an external AFE (gain + ADC) even though their imagers are CMOS. I have seen Analog Devices AFEs in Canon cameras, and expect that these parts can switch gain on pixel-by-pixel basis. This is often needed for converting Bayer array output, and not hard to do with switched cap circuits used in these devices. You don't even need multiple PGA channels, one channel flips gain back and forth at the pixel rate.