In a follow-up to the talk that was previously shared on this blog, here's Digital Camera Myths, Misstatements and Misunderstandings Part II, a presentation by Wayne Prentice to Rochester, NY chapter of IS&T (Society for imaging Science and Tech.) on 17 April. 2024.
00:00 - Introduction
5:51 - Revisiting ISO sensitivity
9:12 - 12 ISO 10/Ha - really independent of camera and illuminant?
13:49 - "It's official: ISO 51,200 is the new 6400". Really?
22:44 - RCCB (Red, clear, clear Blue) sensors yield better SNR. Really?
25:35 - Depth of field: should you always use a longer focal length?
28:18 - sRGB, gamma, CRT display, and Human Vision
31:00 - Questions
This is wrong about RCCB. In Aptina's ClarityPlus technology, which really used a high-transmissivity Y due to not great inter-pixel isolation, we got about a 2-3dB improvement in SNR. The trick Wayne Prentice is missing, is that the noise between color channels is not completely uncorrelated, allowing you to use the C signal to denoise the neighboring R and B.
ReplyDeleteThe color reproduction was shown to be at least as good as a conventional Bayer sensor, both by analysis and in real-world tests.
There were issues that blocked wide scale adoption. One is the non-standard processing, and need for the inter-channel denoise, that made the sensor unsuitable for use with standard phone APs unless an additional pre-processor was used. The other was increased optical crosstalk through the C pixel for light outside the acceptance angle of the microlens. This led to strong purple coloration in off-angle stray light (certain ghosts originating towards the end of the optical path).
I'd say that he is correct about RCCB *under his assumptions*: he uses a traditional demosaic method and SRGB. Aptina developed an pixel processing chain that made better use of RCCB, but I also wonder how well RGGB would work using a better noise pre-processor?
ReplyDeleteOne other major disadvantage of RCCB is dynamic range: the C pixel will saturate much more quickly than the R and B so the RCCB sensor almost requires some kind of HDR design to work in something like natural outdoor illumination. HDR involves even more tradeoffs -- at least until we get well depths closer to ~1M electrons.
Now we should discuss Cyan, Magenta, Yellow!