Instead of the standard RGB three-filter set, our Gemini system uses unique high transmission color filters to produce a full color image from two channels under all lighting conditions, including extreme low light".
Scallop says that from its research over the years in human color vision it has developed a system that human observer perceives at full color despite only 2 colors are being used.
Not much more info is given. A 112MB video file from 1.2MP monochrome CMOS sensor (3.75 µm pixel) equipped with Gemini color filters compares to the color version of the same image sensor.
 |
A video frame recorded at 0.18 lux (equivalent to 3/4 moon), 15 fps taken with a color 1.2 megapixel image sensor. |
 |
| Capture from video - same image as above post-processed to add brightness |
 |
A Gemini prototype video frame taken with the same exposure and frame rate as above, with monochrome 1.2 Mpixel image sensors equipped with Gemini color filters. |
 |
| 0.06 Lux (equivalent to crescent moon), 15fps Capture from Gemini video |
 |
| Magnified image shows there are only two color channels. |
Update: Added another snapshot of Gemini technology demo from the video - the picture below was created out of one monochrome and one filtered channel:
The website has a .pdf download, which while not too much more informative, mentions they are working on an even more sensitive filter scheme with 1 color filter and 1 panchromatic color filter. Is this even possible without horrible noise artifacts?
ReplyDeleteWell, as I already mentioned to the company a few days ago, it is hard to explain the improvement shown on the basis of color filters. Worst case, CFA v. monochrome should yield a 1.7-ish improvement in SNR, and they still use filters so SNR improvement should be less (esp. considering CCM effect). I suspect the conventional color image is just poorly tuned for the imaging conditions.
ReplyDelete@ "even more sensitive filter scheme with 1 color filter and 1 panchromatic color filter."
ReplyDeleteI've updated the post with a screen capture from the video showing this version of image.
Frankly, I find this sort of announcement more annoying than anything considering that Edwin Land did 2-filter color work sixty years ago. If these companies want to be treated seriously, they need to discuss the science.
ReplyDeleteJust stack multiple images :D
ReplyDeleteIs Tenebreax, the parent company, did an add-on to transformer a classic green night vision goggle into a color one ???
ReplyDelete-yang ni
Yes. I didn't realize this but the guys running Tenebraex and making these color devices worked at Polaroid and are using the Retinex function to produce color perception. So it is not like Land, it actually is Land.
ReplyDeleteJust curious, can anyone point me to the reference you are talking about by Edwin Land?
ReplyDeleteScientifc Anerican, May 1959 and December 1977
ReplyDeleteOther references at the end of this article - http://www.neuronresearch.net/vision/files/retinex.htm
More here - http://academic.research.microsoft.com/Author/292945/edwin-h-land
@Dave, thanks a lot for the references.
ReplyDelete