[Digitalis]

why does everybody care so much if this gimmick works on moving subjects or not? I think it's pretty obvious that it doesn't

Because some people have a strong aversion to opinions based on common sense. Who needs common sense when technology can solve everything?

[Fogel70]

IMO it would be more useful if they added semi-automatic stitching feature (using LV to guide stitching),
It would be more flexible (at least for me) and may be used for more types of shots (including added resolution).

It probably requires less R&D resources than pixel shift. And as it's mostly software doing the job, it will not add as much on manufacturing cost or QC.

Input the focal length (or FOV) you want to "simulate" when doing the shot (FI using 35mm lens to create a "12mm" shot), and you start with the center shot and the camera will guide you how to frame next shot with previous shot(s) overlayed on the screen, the camera tell you when the matching is good enough, and when you completed all shots needed. Also for making vertical or horizontal panoramas, and choosing FOV, and what type of projection to use, and maybe even the resolution wanted for the output.

You choose if the stitching should be made in camera (a single jpg output) or later in PP with all separate shots saved as jpg or RAW (or both in camera and PP).

Maybe even Pentax GPS-unit could be used for added functionality (or if the same capabilities are built into the camera) or better guiding of the shots.

[BrianR]

For Pete's sake, why does everybody care so much if this gimmick works on moving subjects or not? I think it's pretty obvious that it doesn't, and why should we care? Exposure stacking doesn't work on moving subjects either, nor do long exposures, that doesn't make them useless features at. It would just be another tool in the box, and a useful one too.

This is pretty much my point, it won't do diddly for anything moving with speed. For more static scenes it seems like a great way to squeeze the most out of the hardware.

[ElJamoquio]

For Pete's sake, why does everybody care so much if this gimmick works on moving subjects or not? I think it's pretty obvious that it doesn't, and why should we care? Exposure stacking doesn't work on moving subjects either, nor do long exposures, that doesn't make them useless features at. It would just be another tool in the box, and a useful one too.

Pentax could intentionally move the sensor during exposure. The downsize is poorer image in pixel peeping, the upside is more color depth?

[tigrebleu]

Using sensor shift to increase resolution through multiple shots isn't new. Hasselblad has a MF digital camera that can take six, 50 Mpix shots and combine them into a single 200 Mpix image by means of shifting its sensor with the help of piezo-electric actuators.

H4D-200MS

The only drawback is that the camera and subject as to be completely still during the shot. The advantages are 1) by shifting up, right, right, down, left and then left (or left, left, up, right, etc., whichever), each photosite can capture all four RGB color components (green is present, which can greatly reduce demosaicing artifacts, and 2) by stacking four images on top one another, noise can be reduced from the signal four times instead of just one. (Unless the noise isn't reduced at all, in which case it would add up with each picture.)

The problem with RGB (or RGBG, to be more accurate) Bayer matrix filters, is that a sensor with 20 million photosites doesn't record each of the 3 primary colors on each of its photosites. Instead, 25% of the photosites record the red color, 25% the blue color and 50% the green color. The missing info is then made up by the in-camera software or the RAW converter software the photographer is using, generating false color, demosaicing artifacts and such.

By moving the sensor six times, the sensor used in the H4D-200MS can record all primary color for each photosite, which results in a theorical resolution of 200 Mpix. Of course, the apparent detail isn't increased as much as a 200 Mpix Bayer matrix sensor would provide. But the colors are much more accurate, and artifacts are next to inexistent. Overall, the visual level of detail of that 200 Mpix picture compares to that a 60 Mpix sensor, but with much, much more accurate colors and much less annoying jaggies and other demosaicing artifacts. But this camera is a specialized product, and was design to answer the need of a very specific market (making photos of works of art and other historical artifacts).

(Just like the latest Sigma Foevon sensor (which is made up of 3 layers of 15+ Mpix that add up to 45+ Mpix) can't beat a 36 Mpix D800 in terms of details, it does provide increase detail, up to that of a 24 Mpix sensor, but more accurate colors, less artifacts and less issues with diffraction because of a too high pixel density. While the D800 36 Mpix roughly has the same pixel density has a D7000 16 Mpix, the 24 Mpix D7100 has a higher density then both the D800 and D7000, which leaves you closer to the limits of diffraction. As a result, diffraction can appear at wider apertures on the D7100, like F/8 instead of F/11. Diffraction reduces sharpness and the level of detail, so the higher sensor resolution can prove counter-productive if shooting at small apertures.)

If Ricoh plans to use such a device to increase resolution, the shift maybe won't provide each photosite with a signal for each primary color, but would at least allow to have two out of the three colors, like RGBG being shifted to GRGB for a RG-GR-BG-GB sampling. That would still require some demosaicing, so the gains wouldn't be as impressive as with a full RBG capture per photosite, both in terms of color accuracy and demosaicing artifacts. But it would allow to increase the level of detail capture by a decent margin at a lower cost.

Some will call this a useless gimmick, other will welcome this added resolution, even if it's limited to still subjects with tripod photography.

Now if a manufacturer can manage to get a one pixel shift in four directions on such a large, 50 Mpix sensor like the one in the Blad, I don't see why they couldn't do it on a roughly 5 times smaller APS-C sensor with 2.5 times less resolution. Doing the maths, that's about 33% more linear total resolution per mm to deal with. A bit more challenging, but nothing that can't be handled, I would assume. Especially considering the shift would only happen in one direction (20 Mpix to 40 Mpix) instead of four (50 Mpix to 200 Mpix).

The question isn't can they do it or not, it's more, how much does it cost to implement such a technology into a camera (in short, how much higher must we sell our camera to get this stuff in, and will the camera's price remain competitive if we do so).

And that is a question I can't answer. But maybe some knowledgeable people know more about industrial processes and how such a device can add to the price of a DSLR.

[Clavius]

Using sensor shift to increase resolution through multiple shots isn't new. Hasselblad has a MF digital camera that can take four, 50 Mpix shots and combine them into a single 200 Mpix image by means of shifting its sensor with the help of piezo-electric actuators.

H4D-200MS

The only drawback is that the camera and subject as to be completely still during the shot. The advantages are 1) by shifting up, right, down and then left (or left, up, right, etc., whichever), each photosite can capture all four RGBG color components, which can greatly reduce demosaicing artifacts, and 2) by stacking four images on top one another, noise can be reduced from the signal four times instead of just one. (Unless the noise isn't reduced at all, in which case it would add up with each picture.)

Now if a manufacturer can manage to get a one pixel shift in four directions on such a large, 50 Mpix sensor, I don't see why they couldn't do it on a smaller (about 5 times), APS-C sensor with 2.5 times less resolution. Doing the maths, that's about 33% more linear total resolution per mm to deal with. A bit more challenging, but nothing that can't be handled, I would assume. Especially considering the shift would only happen in one direction (20 Mpix to 40 Mpix) instead of four (50 Mpix to 200 Mpix).

The question isn't can they do it or not, it's more, how much does it cost to implement such a technology into a camera (in short, how much higher must we sell our camera to get this stuff in, and will the camera's price remain competitive if we do so).

And that is a question I can't answer. But maybe some knowledgeable people know more about industrial processes and how such a device can add to the price of a DSLR.

Excellent post! I think this time we can allow ourselves to get our hopes up and expect a surprise. Remember: Ricoh will want to get their first real offer to the market right. They can't afford to drop the ball during their depute.

[Mehlsack]

Haha, we are at Pentaxforums.com if we are never allowed to raise our hopes because Pentax/Hoya/ricoh will always find a way to not impress us with some super strange features which almost nobody cares or need and the really funny stuff is not to be seen anywhere near pentax

[scratchpaddy]

For Pete's sake, why does everybody care so much if this gimmick works on moving subjects or not? I think it's pretty obvious that it doesn't, and why should we care?

Because cameras are already out there that accomplish the same thing (full color at every pixel) without needing sensor shift, and they do work with moving subjects.

I'm sure this sensor shift thing would add significant cost to the camera. The precision needed to move a sensor exactly one pixel isn't easy to achieve. When Hasselblad came out with the H4D-200MS, they offered to upgrade anyone's old H4D-50MS with the sensor shift for the low, low price of $10,000/€7,000. But maybe I'm wrong. Maybe that's just because it's Hasselblad.

Plus, this is a forum, where we like to talk things over. If something's limited in some way, we're going to talk about it.

[Clavius]

When Hasselblad came out with the H4D-200MS, they offered to upgrade anyone's old H4D-50MS with the sensor shift for the low, low price of $10,000/€7,000. But maybe I'm wrong. Maybe that's just because it's Hasselblad.

Neh, that's not because it's Hasselblad, I bet Ricoh will offer the same service for our existing Pentax DSLRs for the same price.

Who says the technology is needed to shift the sensor exactly one pixel? Maybe the sensor just shifts and the camera takes the second picture the moment it moved one pixel? So no accurate movement, but accurate measuring required. I'm not saying that's the solution, just saying there's much more ways to and end then we think.

[BrianR]

I'm sure this sensor shift thing would add significant cost to the camera. The precision needed to move a sensor exactly one pixel isn't easy to achieve. When Hasselblad came out with the H4D-200MS, they offered to upgrade anyone's old H4D-50MS with the sensor shift for the low, low price of $10,000/€7,000. But maybe I'm wrong. Maybe that's just because it's Hasselblad.

That might not have been a bad deal given the prices of those beasts, the H4D-50MS was €23,000 when in came out in 2009 and the H4D-200MS was €32,000 when it came out in 2011. The prices are closer to each other now, the H4D-50MS is $30,000 vs the H4D-200MS for$36,000 (but if you buy either with photoshop elements 11 at BandH you save $20, whee!!) I dunno if the upgrade would have been worth it, these are definitely above my pay grade.

The multishot difference is pretty impressive, there's a comparison of 100% crops from the different modes in this review: Hasselblad H4D-200MS 200 Megapixel Hands On Preview

[RobA_Oz]

Excellent post! I think this time we can allow ourselves to get our hopes up and expect a surprise. Remember: Ricoh will want to get their first real offer to the market right. They can't afford to drop the ball during their depute.

Exactly so. If what I understand is correct, the appearance of the K-5II/s was an interim measure to give them time to get the next major model release right.

[tigrebleu]

Because cameras are already out there that accomplish the same thing (full color at every pixel) without needing sensor shift, and they do work with moving subjects.

I'm sure this sensor shift thing would add significant cost to the camera. The precision needed to move a sensor exactly one pixel isn't easy to achieve. When Hasselblad came out with the H4D-200MS, they offered to upgrade anyone's old H4D-50MS with the sensor shift for the low, low price of $10,000/€7,000. But maybe I'm wrong. Maybe that's just because it's Hasselblad.

Notice: All the following is to be taken with a grain of salt. I am no expert on manufacturing processes, except for what I read about this subject, so feel free to correct me if I'm wrong.

Yup, it could add additional cost to the camera. But in the case of Hasselblad, the cost to upgrade an old H4D-50MS is most likely higher than the cost to implement the same technology into a H4D-200MS.

Simply put, because they probably need to open the digital back in a white room (so dust doesn't fall on the sensor), install the shifting system without damaging the sensor, make sure the whole thing is up and running, close the digital back and test the device to see if it works properly. Possibly, this operation needs to be done more than once if the first attempt failed, which adds to the man-hours count. Considering the camera probably is assembled in such conditions to begin with, the upgrade probably doubles the cost compared to implementing the system in the first place, as it introduces a new step into the manufacturing process that wasn't there in the beginning, which is to open the digital back, take out the components, upgrade them, put them back in place and close the whole thing, hoping you don't have to do it over again.

The other thing you need to take into account is the complexity of the shifting system and the volume of units produced. The sensor on the H4D-200MS shifts six times (and in all four directions) in order to achieve the 200 Mpix resolution. The sensor on the K-3 might only need to shift one of two times, and maybe just in one direction. This can seriously make the challenge a bit easier for Ricoh engineers.

And while Blad probably had to design a shifting system from the ground up, Ricoh doesn't. They already have a sensor shifting system they know very well, and which improved a lot over the years.

The sensor I have in my K-5 can already shift to correct a small inclination with the horizon level, even when I'm shooting hand held, and while it may not be accurate down to a single pixel, it is accurate enough so I don't have to correct the horizon level on more than 10% of my pictures now. With the K-7 it used to be 90%, for while the K-7 has an horizon level indicator, it's not accurate enough to correct the issue down to below a degree. The K-5 has almost the same horizon level indicator, but you can use the sensor to correct almost any inclination left once the indicator shows a level horizon.

Now regarding sales volume, Hasselblad H4D-200MS sales probably don't count in thousands of units. More likely we're talking about hundreds of maybe even tens of units a year. That makes implementing such a technology at a low price difficult, for the sales never really can make up for all the development required to bring the technology into maturity, so the costs remain higher. The cost of Blad components are also higher to start with, for many of them are manufactured in countries where the workers' income is higher than in the Philippines, China or Malaysia. The cameras are also assembled mostly by hand, whereas DSLRs tend to require less hand-made assembly operations.

This helps keep the cost of DSLRs low, and the huge amount of units produced help absorb the costs of R&D investments in a rather rapid manner. Hasselblad can't do the same here, for their sales volumes aren't even close to those of those of the D4 or 1DX, which remain speciality products in the realm of the DSLR, let alone cheaper DSLRs like the K-5, D7100 or EOS 70D.

(The D800 was produced at around 30,000 units per months in the early months it was introduced, which helped keep the cost down, and it was in fact more than two times less expensive than the D3X, yet with better AF, higher resolution, higher dynamic range, better high ISO noise levels, etc., with the D800 being just a bit less rugged than the D3X in the end. All this thanks to more—much more—units being manufactured than with the D3X.)

Now Ricoh could manage to keep the costs of a pixel-shifting device low enough by spreading it over a bigger volume of DSLR sales, and if the system is simpler to begin with (less shifting), I suppose it can remain affordable enough to keep the K-3's price close to that of the K-5 II's. In fact, besides the AF system and the sensor (and attached A/D converter and pipeline), I'd bet not much else will change in the upcoming K-3. The body could be the same as the K-7/K-5/K-5 II (and IMHO, it should be the same), the LCD monitor, viewfinder and the like don't need to be changed (except for the focusing screen, if the new AF has more focus points), the memory card slot will probably be the same as well, etc.

By now, all the manufacturing infrastructure needed to produce these components has already been paid for (at least for the most part), so the costs are limited to the R&D investments, the raw materials needed to produce the components and the workforce needed to assemble them.

But all this is theory indeed. I am no expert on the matter, but it's my guess.

[Col]

Haha, we are at Pentaxforums.com if we are never allowed to raise our hopes because Pentax/Hoya/ricoh will always find a way to not impress us with some super strange features which almost nobody cares or need and the really funny stuff is not to be seen anywhere near pentax

Yeah what's with this colour stuff

[Kunzite]

I doubt we'll see this feature in the next Pentax camera(s), but there's something you're forgetting: the SR mechanism is actually quite precise. Its AstroTracer function, for example, can follow the stars movements for up to 5 minutes.

[mamethot]

The sensor glide hardware is one thing, software to operate it is another. Both are needed. Probably a lot of software engineering is required to make something useful out of this hardware.