Friday, January 17, 2014
Fujifilm Proposes RGBW Pixels with Different Size
Japan-language Egami blog reports that Fujifilm applies for a Japanese patent 2013-258168 on different-sized RGBW filter pattern. The green and white pixels are bigger to minimize the luminance noise at the expense of chrominance noise elevation. YourNewsTicker nicely painted the B&W figure from the application:
6,137,100 claims seem to cover this...
ReplyDeleteThose big white (and green) pixels are going to saturate well before the little red and blue ones. Does this not present a problem?
ReplyDeleteI think the this is problem as White &Green need large FWC. Maybe photodiode and FD are differenent
DeleteIn a way that's what their going for. The white and green will "fill up" faster, therefore will need to be amplified less in low light situations. The red and blue will simply be amplified, increasing chromatic noise.
DeleteSince saturation will occurr in green and white, you might even be able to reconstruct the colour of the saturated parts from blue and red, if they are not saturated. The details will be lost with green and white though. This is somewhat speculation though.
Pixel size should not effect the saturation since both the incident light and the well capacity are proportional to pixel area, all else being equal.
DeleteAs for clipped highlights, you can recover them on RGBW patterns such as those from Truesense and Sony. Once W pixels are saturated, you can fall back on RGB only demosaicking. For example see http://imagealgorithmics.com/rgbwhdr.html
-Tripurari
When saying they would saturate first, I was going under the assumption that saturation would be limited by the floating diffusion. Are they different for the G & W pixels?
Deletethe photodiode should be larger, or multiple ones of the same size, for green but white will need special treatment.
DeleteI thought the smaller ones would saturate faster?
ReplyDeleteAren't humans more sensitive to chrominance noise?
ReplyDeleteThe different sized pixels gets around the problem of G or W pixels saturating first (if they were all the same as they are in the RGBW layout that would be an issue).
ReplyDeleteAnd no this is a novel patent. Those other RGBW patents don't have different sized pixels.
I note that the pixel array is really 2 x 3 for each pixel. If you made those the W and G out of two subpixels and put a single microlens over them then you'd be able to do PDAF ... just like Canon!
And each quarter pixel rotates by 90 degress as you go around the "bigger" pixel. Just like Xtrans. Of course non rectangular sensels will make demoasicing, err, interesting.
so we see 4 color filters with 4 x 3 = 12 photodiodes behind (4G + 4W + 2B + 2R). there could be more complex patterns like some cells with deeper wells or darker filters for high light. but as the complexity goes up the technology will go finer all the way towards mobile phones.
DeletePixel based exposure they r planning for?
ReplyDeleteXPro2 sensor?
ReplyDelete