[Usuqa]

We know that we can shift the sensor in the k5 manually and I have done it quite a few times.. its a great feature .
We also know that we can attach a gps and have the camera track stars using the sensor shift..
The camera uses sensor shift to correct for shake automatically.
The camera can use sensor shift/rotation to auto correct horizon levels.

Couldn't the sensor shift also be used to capture mode detail scenes in still life?, The HB200 shifts its sensor 1 1/2 pixels and shoots 6 images to create a 200mp file from the 50mp sensor.. this is something that pentax sensor shift would be able to do (hell it tracks stars) and would be a great feature to have for people shooting in a studio.

[Anvh]

I'm curious why they didn't put it in the camera, they might still do it with a firmware update but i doubt that...

You can do the same manual with sensorshift but it takes a bit more effort and time.

[Wheatfield]

They can only build so much stuff into these things, and they like to leave some things not done so that they can put stuff into later models.
In this case, odds are, they either didn't think of it, or didn't think the market needed it.

[alohadave]

Couldn't the sensor shift also be used to capture mode detail scenes in still life?, The HB200 shifts its sensor 1 1/2 pixels and shoots 6 images to create a 200mp file from the 50mp sensor.. this is something that pentax sensor shift would be able to do (hell it tracks stars) and would be a great feature to have for people shooting in a studio.

How much improvement are you going to get by adding 1.5 pixels each direction?

[Anvh]

How much improvement are you going to get by adding 1.5 pixels each direction?

The hassie moves in a square i believe taking 4 photos so the gain in pixels and sharpness is against 4 times tops.
i just thought that if they move the sensor 3 times stacking red green and blue pixels ontop of eachother they can build up the pixels without demosaicing.
You get the same effect as feavon sensor for example, pixel count stays the same but the quality and accuracy of the pixels improves alot.

[Gimbal]

How much improvement are you going to get by adding 1.5 pixels each direction?

No, they move 0.5 pixels and take a new shot, thus doubling the resolution in that direction. Do that 4 times and we would get a 48Mpixels picture. The question is if the SR system can handle it.

[Usuqa]

No, they move 0.5 pixels and take a new shot, thus doubling the resolution in that direction. Do that 4 times and we would get a 48Mpixels picture. The question is if the SR system can handle it.

Ahh ok the article I read said 1 1/2 a pixel but then I didn't read many other's , If the SR can handle trailing stars with a gps unit I think (but could be way wrong) that shifting slightly while on a tripod for this sort of thing should be pretty easy.

[zapp]

48 MP images are not very handy - I hate my 120 MB raw files from a 56MP back. The Hasselblad 200 MP are a marketing gag - Toshiba offered this feature in a 6 MP camera 10 years ago.

A simple shift will at least double the pixel count. While Hasselblad does not use an anti alias filter, Pentax does. Pentax pixels are much smaller than Hasselblad pixels. So, I assume that we can not expect better resolution from the Pentax by shifting the sensor - I doubt that this really works for the Hasselbald solution either. The whole feature only works from a tripod and will not work during live view. It would be interesting to see if the Hasselblad solution improves RGB sampling of the images. Maybe we could get better monochrome from this procedure or less colour aliasing (something that only works without an anti-alias filter). Moving 0.5 pixel will result in a new image where 50% of the information was already sampled in the first image by the same pixel. Moving 1.5 pixels allows for a more independent sampling.
Stitching solution are actually more staright forward - requiring much larger sensor shift. In this case you really gain resolution. This is a pain with a camera where lens and sensor a fixed, a joy with technical cameras allowing for X/Y movement, i.e. Alpa.

[Gimbal]

Moving 0.5 pixel will result in a new image where 50% of the information was already sampled in the first image by the same pixel. Moving 1.5 pixels allows for a more independent sampling.

How so? The sampled picture would be exactly the same as the one shifted 0,5 pixels, except for the 1 pixel shift.

[Anvh]

How so? The sampled picture would be exactly the same as the one shifted 0,5 pixels, except for the 1 pixel shift.

To create 1 pixel, information of the surrounding pixels is also use.
If you've for example a green pixels it needs the info of the surrounding blue and red pixels to make one RGB pixel.
A shift of 0,5 pixels will mean that roughyl the same information is used to creat the same pixel, it's indeed better to shift 1,5 pixels.
Or you shif 1 pixel so that the blue pixel next to the green will be in the spot of the green pixel and then move the sensor again to move a red pixel in the same place, this would yield not in a larger file but it will give you a more accurate pixels because it doesn't need to mix the surrounding pixels.

[Gimbal]

A shift of 0,5 pixels will mean that roughyl the same information is used to creat the same pixel,

Yes, isn't that the whole idea? Say that there is a very small detail (black dot on white backround) that on your first shot happens to fall right between two pixels, creating two gray pixels. Shift 0,5 pixels and shoot again, this time the result is one white and one black pixel (slightly simplified). Add this together and we get white, grey, black, grey instead of the original two grey pixels.

Then, as you say, we have color filters and demosaic algorithms, maybe that changes things, but my gut feeling is that is doesn't.

Anyhow, wasn't Ricoh one of the first cameras introducing this trick some 12 years ago? With the Ricoh RDC-7.

[zapp]

Thank you Anvh...

Imagine a typical Bayer pattern of pixels in one row: R-G-B-G-R-G-B
now move the sensor by 0.5 pixels ........................... -R-G-B-G-R-G-B
now move the sensor 1.5 pixels .................................... -R-G-B-G-R-G-B

case pink: the new pixel cover to one half the same information as before and to one half new information, i.e. the first R pixel cover 50% R and 50% G signal
case green: the new pixels cover up to 100% new color information - the first R pixel falls between a G-B pixel thus covering 100% new information.


While all this is true, the Bayer pattern demosaics the single layer R-G-B file to a three layer file with 100% R, 100% B and 100% G coverage. The normal sensor uses interpolation to do this, the shifted sensor adds phyical information to fill the gaps.

I would actually prefer the single shot Foveon sensor for full colour info. A multishot sensor shift technique may also be used to create a much cleaner image while keeping the resolution the same.

[Anvh]

Yes, isn't that the whole idea? Say that there is a very small detail (black dot on white backround) that on your first shot happens to fall right between two pixels, creating two gray pixels. Shift 0,5 pixels and shoot again, this time the result is one white and one black pixel (slightly simplified). Add this together and we get white, grey, black, grey instead of the original two grey pixels.

For grey scale your method works indeed very well but with colours it's better to shift more.
If you shift 1,5 pixels you've the plusside you say plus more accurate colours and even more details.

[falconeye]

We know that we can shift the sensor in the k5 manually and I have done it quite a few times.. its a great feature .
We also know that we can attach a gps and have the camera track stars using the sensor shift..
The camera uses sensor shift to correct for shake automatically.
The camera can use sensor shift/rotation to auto correct horizon levels.

Couldn't the sensor shift also be used to capture mode detail scenes in still life?, The HB200 shifts its sensor 1 1/2 pixels and shoots 6 images to create a 200mp file from the 50mp sensor.. this is something that pentax sensor shift would be able to do (hell it tracks stars) and would be a great feature to have for people shooting in a studio.

This is called superresolution and you already got it: shoot a burst of say 7 photos with SR on (tripod ok) and use PhotoAcute to create the superresolved photo from them. The trick with SR is that you then don't get 7 pixel-identical images. And that's all it takes.

[Usuqa]

This is called superresolution and you already got it: shoot a burst of say 7 photos with SR on (tripod ok) and use PhotoAcute to create the superresolved photo from them. The trick with SR is that you then don't get 7 pixel-identical images. And that's all it takes.

Nice sounds like something to try