[pinholecam]

Didn't Pentax in their early pixel shift presentations mention that there were other algorithms/methods available, but they created very large files and needed more shots to be worth it?

[photoptimist]

So if im reading this right, would adding more pixels on the sensor give you a better image then using pixelshift? Dont have to worry about movement, also lets say if we take each pixel and either dividing each pixel in half or quarters would give you roughly the same as shifting the sensor because the camera is having smaller sections to sample for the different colors(essentially a 72 mp without pixelshift version of the k-1 vs a k-1 with pixelshift) ? In this case im ignoring the increase of noise in less then ideal lighting due to pixel size.

In one-shot mode, a Bayer-filter camera measures only 25% of its pixels in red, 50% in green, and 25% in blue. (Digital zoom then interpolates these sparse measurements to fill in all the missing data.)

So the K-1 (and every other 36 MPix Bayer filter camera on the market*) really is only a 9 MPix red-sensitive camera, 18 MPix green-sensitive camera, and 9 MPix blue-sensitive camera.

In 4-shot pixel shift mode, the K-1 becomes a 36 MPix red-sensitive camera, 36 MPix green-sensitive camera, and 36 MPix blue-sensitive camera. In addition, the green channel gets measured twice so noise in the green channel drops.

A 72 MPix Bayer-filter camera would replicate the green-channel resolution of a K-1 in pixel shift but have poorer resolution in red and blue.

A 144 MPix Bayer-filter camera would replicate the red-channel and blue-channel resolution of a K-1 in pixel shift but have higher resolution in green.

Overall, one would need about a 108 MPix sensor to get the same average resolution as the K-1 in pixelshift mode.

(*Note: Fuji's XTrans and Sigma's Foveon are different kettles of fish with other resolution and color accuracy issues.)

[Fcsnt54]

In one-shot mode, a Bayer-filter camera measures only 25% of its pixels in red, 50% in green, and 25% in blue. (Digital zoom then interpolates these sparse measurements to fill in all the missing data.)



So the K-1 (and every other 36 MPix Bayer filter camera on the market*) really is only a 9 MPix red-sensitive camera, 18 MPix green-sensitive camera, and 9 MPix blue-sensitive camera.



In 4-shot pixel shift mode, the K-1 becomes a 36 MPix red-sensitive camera, 36 MPix green-sensitive camera, and 36 MPix blue-sensitive camera. In addition, the green channel gets measured twice so noise in the green channel drops.



A 72 MPix Bayer-filter camera would replicate the green-channel resolution of a K-1 in pixel shift but have poorer resolution in red and blue.



A 144 MPix Bayer-filter camera would replicate the red-channel and blue-channel resolution of a K-1 in pixel shift but have higher resolution in green.



Overall, one would need about a 108 MPix sensor to get the same average resolution as the K-1 in pixelshift mode.



(*Note: Fuji's XTrans and Sigma's Foveon are different kettles of fish with other resolution and color accuracy issues.)

Ok so essentially you need the shift to get the same percentage across each color channel. Although by cutting each k-1 sensors pixels into fourths you are only increasing resolution and not color accuracy. Would it matter as much saying you are sampling smaller areas though? I mean yes its 25%, 25%, and 50% but thats still a pretty decent percentage with the much smaller pixels? Granted low light would suffer unless the manufacturers work their magic...

Another question, did film have the same issue with percentage with reading red greens and blues, or is it something that we are now worry about with digital. And because we are only having an issue with digital, they use this high res/pixelshift to compensate to bring it back to the level of film?

[photoptimist]

Ok so essentially you need the shift to get the same percentage across each color channel. Although by cutting each k-1 sensors pixels into fourths you are only increasing resolution and not color accuracy. Would it matter as much saying you are sampling smaller areas though? I mean yes its 25%, 25%, and 50% but thats still a pretty decent percentage with the much smaller pixels? Granted low light would suffer unless the manufacturers work their magic...

Another question, did film have the same issue with percentage with reading red greens and blues, or is it something that we are now worry about with digital. And because we are only having an issue with digital, they use this high res/pixelshift to compensate to bring it back to the level of film?

Although you could call it color accuracy, it's really more a matter of color resolution. If you illuminate a standard resolution test target in pure red light, you'll find that all Bayer filter cameras seem to have 1/2 the lpm resolution (or 1/4 the effective number of pixels) that you'd expect. Worse, if the lens is sharp and the sensor lacks a low-pass filter (or the anti-aliasing simulator in the newer Pentax cameras), then a thin pure red line on a black background can disappear if the image of the line happens to miss the red-sensitive pixels. The same can happen with blue but it's much less likely to happen with green. High resolution black-and-white objects may fall in strange ways across the color-sensitive pixels creating those strange false colors of aliasing or moire.

As to your second question, no, film did not have this problem and, yes, pixel shift is a hack to correct for the problems of the Bayer filter. Because of the way color emulsions were constructed as a stacked series of color sensitive layers, every little part of a piece of color film was sensitive to every color. However, color film was limited in it's resolution relative to black-and-white films because the color dye clouds in the developed film were about 10X the size of the light-sensitive silver grains. The effective pixel sizes for even the best low-ISO color films were larger than what current day sensors offer and the high-ISO color films were low-res horror shows compared to what the K-1 can do at ISO 1600. (That said, there are other differences between film and digital that digital has yet to solve.)

The closest digital technology to film (besides pixel shift) is the Sigma Foveon sensor which cleverly uses a property of silicon whereby the upper surface layers of the chip tend to get more of the blue light signal and the deeper layers of the chip tend to get more of the red light signal. By measured pixel values at three depths and doing a bit of math, Foveon can estimate red , green, and blue simultaneously at every pixel. Unfortunately, the Foveon's color separation is not as well controlled as the color filters of a Bayer filter camera and the math needed to estimate color makes the image much noisier.

---------- Post added 01-18-18 at 07:36 PM ----------

P.S. If you want to read a more technical report about film grain and resolution, see http://cool.conservation-us.org/coolaic/sg/emg/library/pdf/vitale/2007-04-vi...resolution.pdf

[MikeyBugs95]

I can see an industrial use for this camera for detecting physical defects or product photography. Other than that, and because it's so late, I don't know.

[Fcsnt54]

Although you could call it color accuracy, it's really more a matter of color resolution. If you illuminate a standard resolution test target in pure red light, you'll find that all Bayer filter cameras seem to have 1/2 the lpm resolution (or 1/4 the effective number of pixels) that you'd expect. Worse, if the lens is sharp and the sensor lacks a low-pass filter (or the anti-aliasing simulator in the newer Pentax cameras), then a thin pure red line on a black background can disappear if the image of the line happens to miss the red-sensitive pixels. The same can happen with blue but it's much less likely to happen with green. High resolution black-and-white objects may fall in strange ways across the color-sensitive pixels creating those strange false colors of aliasing or moire.



As to your second question, no, film did not have this problem and, yes, pixel shift is a hack to correct for the problems of the Bayer filter. Because of the way color emulsions were constructed as a stacked series of color sensitive layers, every little part of a piece of color film was sensitive to every color. However, color film was limited in it's resolution relative to black-and-white films because the color dye clouds in the developed film were about 10X the size of the light-sensitive silver grains. The effective pixel sizes for even the best low-ISO color films were larger than what current day sensors offer and the high-ISO color films were low-res horror shows compared to what the K-1 can do at ISO 1600. (That said, there are other differences between film and digital that digital has yet to solve.)



The closest digital technology to film (besides pixel shift) is the Sigma Foveon sensor which cleverly uses a property of silicon whereby the upper surface layers of the chip tend to get more of the blue light signal and the deeper layers of the chip tend to get more of the red light signal. By measured pixel values at three depths and doing a bit of math, Foveon can estimate red , green, and blue simultaneously at every pixel. Unfortunately, the Foveon's color separation is not as well controlled as the color filters of a Bayer filter camera and the math needed to estimate color makes the image much noisier.

---------- Post added 01-18-18 at 07:36 PM ----------


P.S. If you want to read a more technical report about film grain and resolution, see http://cool.conservation-us.org/coolaic/sg/emg/library/pdf/vitale/2007-04-vi...resolution.pdf

Thankyou. Looks like it will be a good read.