[falconeye]

I'm pretty sure this shake-AA system will also be helpless in video. Any AA will have to be done in software.

I beg to differ and it depends.

I agree that Pentax will most likely have dismissed video. But not for technical reasons.

First and to clarify, video created from downsampled (aka supersampled) frames using all sensor pixels have no aliasing problem and almost no visible Bayer-moiré problem, esp. with a still image Bayer-AA filter at work.

The problem is that most dSLRs (and I assume the K-3 will be no exception) use subsampling, i.e., they skip pixels or entire lines when creating the video signal. This creates ordinary moiré artefacts rather than color moiré artefacts a Bayer-AA filter is designed to avoid. This is an ugly artefact esp. in video as it causes nasty edge flickr. It renders a camera useless for professional use (except if there is a mirror box accessory like for the D800).

The sensor-shift AA can be reconfigured to cure video aliasing as well and I already mentioned it when I proposed the idea back then: Assume you need to scale the video signal down to 2.7MP to preserve a little headroom to downsample HD. Thats 1/9th #pixels or 1/3rd pixel size in video. The vibration amplitude would need to be 6 microns rather than 2 microns. But at the same time, video exposure times are slow, say 1/100s, 1/48s or less. So, let the vibration operate at 1/2 the frequency, i.e., 250Hz and you end up with 3/4 the acceleration of the Bayer-AA mode. The video-AA mode is actually technically less demanding, assuming a 6 micron shift is still tiny which it is. In video mode, the motion would be a bit more complex to draw a figure more complex than a single ellipse. But no problem as 250Hz yields enough time to sweep multiple times.

So no, not ANY AA for video has to be done in software. If Pentax is clever enough, they end this. Unfortunately of course, at 500Hz or 250Hz, the AA is audible. Which is why I looked at ultrasonic vibration frequencies.

---

Last edited by falconeye; 10-08-2013 at 06:09 PM.

[JimmyDranox]

Falconeye, maybe you're right, maybe not. I don't know if we have enough information about how the system works. We will see.

[asp1880]

Moreover, the prime task of said content analysis is to track the subject (3D tracking, no idea if the K-3 features it). Frame by frame, the firmware looks for the subject to be tracked (the point the AF.C locked on initially) and reports the image coordinate to the phase AF module which, in turn, determines if any of its AF points covers it. If yes, AF will be adjusted. If no, the AF stays idle (or uses prediction) until the subject moves back into the region covered by AF sensors. This enables the camera to track a subject even if It moves out of the AF area. In practice due to the low resolution, this works well if the tracked subject is well isolated against the background, e.g., by color.

The videos on the Ricoh website seem to suggest that the K-3 does Nikon-style "3D tracking".

Eventually, it will allow the camera to detect everyone's smile and trigger automatically.

Or just have the camera detect and expose for the face. D800 can also do this.

Regards,
--Anders.

[Frater]

If someone was drawing a chart showing AA effectiveness per shutter speed, then I would expect to see ripples getting bigger with faster shutter speeds.

For example at 1/125, 1/250, 1/500, the curve would be at the same peak levels.
But for 1/350, I would expect a low point of the curve, as the SR effectiveness would be only half of the value at 1/500. This would be because one side of the SR circle got double exposure compared the other side, so that the contrast of the moire artefact is only halved, but not removed. But this would depend on the orientation of the photographed structure, so we are actually considering "worst cases".

This is a slight natural disadvantage of SR for AA, unless they change the rotation speed depending on the shutter speed (which I don't know, but which would make sense, and would even help in saving some battery power?)
Falk, please purchase a K-3 and play around with it

---

Last edited by Frater; 10-09-2013 at 01:27 AM.

[falconeye]

If someone was drawing a chart showing AA effectiveness per shutter speed, then I would expect to see ripples getting bigger with faster shutter speeds.

For example at 1/125, 1/250, 1/500, the curve would be at the same peak levels.
But for 1/350, I would expect a low point of the curve, as the SR effectiveness would be only half of the value at 1/500.
...
Falk, please purchase a K-3 and play around with it

You are technically correct about the ripples effect when drawing a graph.
However, the dips won't be as deep. It is correct that at, e.g., 1/333s, the blurring isn't perfect with 1.5 waves occuring. Depending on phase (image location), the blurring of pixels is 2:1 (bad) or 1.5:1.5 (perfect). But even with the worse 2:1 blurring, there will be typically enough smeared color information left to avoid color moiré when reconstructing rgb color during demosaicing. Therefore, I think the ripples effect won't remain visible when drawing color moiré amplitude vs. shutter speed.

Wrt purchasing a K-3. If working as advertized, I think the K-3 is an almost perfect camera in a 1 camera setup scenario. However, I now use several cameras incl. one FF and for me personally, the K-3 offers little advantages over my K-5 where I now use it. Esp. as it got heavier than a D7100 and now is almost same weight as a D610.

[falconeye]

Falconeye, maybe you're right, maybe not. I don't know if we have enough information about how the system works. We will see.

Maybe I am wrong. But we won't see as I am sure Pentax didn't implement any of the ideas laid out above.

[Iberia]

I wonder if that "microscopic" blur will have a negative impact on SR performace?

And I also wonder if the use of a mechanical AA won't reduce the SR lifetime in the long run?

R

[Ex Finn.]

I wonder if that "microscopic" blur will have a negative impact on SR performace?

And I also wonder if the use of a mechanical AA won't reduce the SR lifetime in the long run?

R

I am not an expert but I think that the shutter will die long before the SR.

[falconeye]

Digitalis made a point I forgot to mention ( https://www.pentaxforums.com/forums/pentax-k-3/239081-k-3-finally-has-what-iv...ml#post2533747 ):

The AA feature won't work with flash of course, i.e., won't work in the studio.

It is in the studio with fabric where I typically see moiré with my D800E, in between 0% and up to 10% of shots, depending on the session (normally treatable in post). That's a BIG caveat and lets me predict other vendors won't copy the idea. Another reason to make it work at ultrasonic speed.

[baro-nite]

Digitalis made a point I forgot to mention ( The K-3 finally has what I've been looking for in a DSLR. ):

The AA feature won't work with flash of course, i.e., won't work in the studio.

I had been wondering about that myself. Good to hear you normally find this treatable in post; perhaps the K-3 will be (nearly) as amenable to such treatment?

[Class A]

But even with the worse 2:1 blurring, there will be typically enough smeared color information left to avoid color moiré when reconstructing rgb color during demosaicing. Therefore, I think the ripples effect won't remain visible when drawing color moiré amplitude vs. shutter speed.

What concerns me, though, is the non-uniformity of the smearing.

A standard Bayer-AA-filter provides a uniform smearing that can be theoretically undone through deconvolution. I don't see anyone developing anything close to an optimal reconstruction of the signal the K-3 AA simulator turned into non-uniform smears.

While the general SR-based AA-simulator idea is great, the current implementation in the K-3 appears to be lacking.

I'm wondering whether the ultra-sonic actuator for the sensor cleaning could be used in any way. Probably the amplitude achievable with it isn't sufficient.

The AA feature won't work with flash of course, i.e., won't work in the studio.

Depends on the flash duration.

Portable speedlights typically have a T.1 time around 1/650s at 1/2 power and even 1/125s at full power.

Studio strobes all appear to be too fast.

I wished the K-3 were available with a conventional Bayer-AA-filter as well.

[WerTicus]

Exactly. It's all about peace of mind. Without the AA filter on the K-5 IIs, a lot of people thought "OMG!!! Moire is going to ruin all of my pictures!!!"

Yes I looked at how hard it is to remove moire in post and decided it was not worth it.

[WerTicus]

I am not an expert but I think that the shutter will die long before the SR.

The SR died in my K20d before the shutter... Was under warranty too so not a big deal ;-)

[falconeye]

What concerns me, though, is the non-uniformity of the smearing.

That's a point of concern, of course.
But I assume that for all speeds except 1/1000s, the non-uniformity will be very hard to notice, esp. in circular mode. Even at 1/1000s and in circular mode where half a circle would have a similiar smearing effect independent of orientation, the effect may only be noticeable at longer than 20% image height vertical high contrast edges.

[Adam]

Yes I looked at how hard it is to remove moire in post and decided it was not worth it.

It's actually really simple. Just open your JPEGs in camera raw (or if you shoot raw, go straight to the moire removal slider) and you'll be good to go. The Ricoh GR even has an in-camera moire removal feature that works fairly well.

PP is not a replacement for an AA filter in my opinion, but it can still save your photos.