combatentropy
Veteran
Sony announced a new sensor pattern. They call it Q67.
All they did was rotate the Bayer pattern 45 degrees. But this elegant solution brings two advantages:
1. Debayering is far easier. Classic debayering is like wine: you need to spend a lot of time to get a good result. This is why Ikonoskop and the Drama don't even do it in camera. They record an uncompressed, raw signal from the sensor and let you do at in the leisure of your editing studio.
The classic Bayer pattern is, I think, clever. At first you may think that it holds only 1/3 resolution for each color. But because of the the photosites are staggered, it is possible to deduce more. (Something like, this green pixel is dark and surrounded by bright red pixels. Therefore red light is falling on this green pixel, too.) So there's a lot of information in there, but it's complicated to unravel it.
With the new Q67 pattern, each pixel has one green pixel, two blue slices, and two red slices. Deducing the color at each pixel is a lot easier: The green channel is the value of the green pixel. The blue channel is the average of the values of the two blue triangles. Likewise with the red. I imagine that even the Ikonoskop, with its design goal to make a low-power, fanless camera, could do this in camera.
2. If you can debayer in camera, you could take advantage of the many visually lossless compressions that operate on YCrCb colorspace, like ProRes or DNxHD.
You can compress raw Bayer, but it's hard. Most of the established codecs work on YCrCb images. The codecs for Bayer that I've seen all involve wavelets --- which are cool, but they seem to take a lot more power.
With Q67, I wonder if you could debayer and compress with even a small ARM processor. So you could keep the camera simple and fanless (like the Ikonoskop) but still use off-the-shelf cards (like CompactFlash).
All they did was rotate the Bayer pattern 45 degrees. But this elegant solution brings two advantages:
1. Debayering is far easier. Classic debayering is like wine: you need to spend a lot of time to get a good result. This is why Ikonoskop and the Drama don't even do it in camera. They record an uncompressed, raw signal from the sensor and let you do at in the leisure of your editing studio.
The classic Bayer pattern is, I think, clever. At first you may think that it holds only 1/3 resolution for each color. But because of the the photosites are staggered, it is possible to deduce more. (Something like, this green pixel is dark and surrounded by bright red pixels. Therefore red light is falling on this green pixel, too.) So there's a lot of information in there, but it's complicated to unravel it.
With the new Q67 pattern, each pixel has one green pixel, two blue slices, and two red slices. Deducing the color at each pixel is a lot easier: The green channel is the value of the green pixel. The blue channel is the average of the values of the two blue triangles. Likewise with the red. I imagine that even the Ikonoskop, with its design goal to make a low-power, fanless camera, could do this in camera.
2. If you can debayer in camera, you could take advantage of the many visually lossless compressions that operate on YCrCb colorspace, like ProRes or DNxHD.
You can compress raw Bayer, but it's hard. Most of the established codecs work on YCrCb images. The codecs for Bayer that I've seen all involve wavelets --- which are cool, but they seem to take a lot more power.
With Q67, I wonder if you could debayer and compress with even a small ARM processor. So you could keep the camera simple and fanless (like the Ikonoskop) but still use off-the-shelf cards (like CompactFlash).
Last edited:
