Zone plates for FS700 and FS100

[Matt Davis]

Originally Posted by David Heath

I put that down to them having the same sensor, but maybe the FS700 having a more aggressive OLPF than the FS100??

Originally Posted by Alister Chapman

poor processing can also lead to aliasing and resolution loss.

Whilst discussing this FS700 camera with Sony (not covered by NDA), there was a strong view that this would be a B-Cam to the F3, whilst the FS100 would be a B-Cam to the FS700 - but only at a lower tier of priority.

So, the FS100 has the same sensor as the F3 but without the processor, the FS700 has a different sensor but emulates the same Gamma settings as the F3 but with less finesse and processing power to enable better power usage and leaving a little wiggle room for a 4K F5 perhaps using the SAME sensor as the FS700 but with its full-on image processor and S-Log.

One can sympathise with the 'I don't shoot charts' group of camera testers... Those charts begin to look like entrails after a while (in the haruspicific sense).


BTW - Point of Order: Mr Chapman, does the F3 switch off all Knee functions if a CineGamma is selected, like the EX1? I found out to my chagrin that you can have both knee AND CineGamma 3 & 4 on at the same time, which prevents the latter from stepping over 100 IRE.

[Felix Steinhardt]

What about this FS700 chart?
http://provideocoalition.com/images/...700-2K-024.jpg

The diagonal lines go clearly beyond 1000 but horizontally moire and false detail starts to show from 800.

[David Heath]

Originally Posted by Felix Steinhardt

What about this FS700 chart?

The diagonal lines go clearly beyond 1000 but horizontally moire and false detail starts to show from 800.

It's normal. In a straightforward case of hor/vert columns and rows of photosites, the resolution of diagonal lines will be up to 1.4x (sq rt 2) the resolution of horizontal or vertical lines, and it's pretty easy to prove mathematically why that will be the case.

If the fundamental limit of horizontal/vertical res is due to the row/column spacing, the limiting factor for the diagonal res will be lines drawn corner to corner across the phosites at 45deg. If you draw it out, and do the geometry, you'll find the lines are more closely spaced than the photosite spacing - hence the up to 1.4x figure.

Practically, a lot may depend on things like the blank zones between photosites - it's wrong to think of them as like the squares on graph paper (closely butting one against the other). You have to think of the graph paper with the lines quite thick compared to the square-square spacing.

Shift the rows on to the 45deg axis and the whole picture moves. The fundamental row/column resolution is now corresponding to diagonal lines, with the hor/vert resolution up to 1.4x greater. Just look at Alan Roberts chart for the Z5/Z7 and that's clear from his zone plate.

[Alister Chapman]

But the Z5/Z7 is a 3 chip camera. There are no gaps between the green red or blue samples as with a single chip camera, so you can't use that as an example of how a single sensor clearvid camera will perform.

I think David your muddling 3 chip or monochrome and single chip designs. You talk of diagonal resolution being greater, which it is on a 3 chip or monochrome sensor camera. But it is not on a bayer camera as there is a gap between green samples on the diagonal axis that is occupied by the red and blue pixels, on a bayer sensor the diagonal resolution is 0.707 x H. Clearvid CFA is not as simple as as a rotated sensor because the sensor is still read horizontally and the pixels are now offset from each other by only half a pixel as opposed to a full pixel and 4x green pixel interpolation is used to very accurately calculate the in between values.

[David Heath]

Originally Posted by Alister Chapman

I think David your muddling 3 chip or monochrome and single chip designs.

No - it's true I was making the point without reference to colour (for simplicity), but the point holds good no matter what. It's easy enough to see it with monochrome or 3 chip, less intuitive with a Bayer, but the theory still holds good.

You talk of diagonal resolution being greater, which it is on a 3 chip or monochrome sensor camera. But it is not on a bayer camera as there is a gap between green samples on the diagonal axis that is occupied by the red and blue pixels, on a bayer sensor the diagonal resolution is 0.707 x H. .

Ah, but you could say there are gaps between the green samples on the horizontal and vertical axes by the same logic. It would then follow that the horizontal resolution is only 0.5 what it would be for a monochrome sensor - amd yes, 0.707 for diagonal resolution. In other words, diagonal resolution is then 1.4x horizontal.

Practically, all that ignores deBayering. It also ignores that for a normal deBayered sensor the aliasing will be coloured, and will differ with orientation. For horizontal and vertical lines, if the sensor receives white on one row, black on the next, it must follow that only red/green OR blue/green photosites are stimulated - hence expect yellow/cyan aliasing. For diagonal lines, expect a similar situation with spatial frequencies 1.4x higher - in which case either diagonal rows of green only OR red/blue photosites are stimulated - hence expect green/magenta aliasing.

That gets illustrated beautifully in a zone plate by Alan Roberts of the F3 - http://downloads.bbc.co.uk/rd/pubs/w...ony_PMW-F3.pdf - and scroll down to figure 2. Ignore the red/green zone plates - look at the small b/w chart centre top. Yellow/cyan aliasing on the hor/vert axes, green/magenta on the diagonal. And the green/magenta nulls are directly above/below and to the side of the yellow/cyan nulls. Geometry shows they must therefore be sq rt 2 times as far from the centre (1.4), so must be for a spatial frequency 1.4x greater. Exactly as predicted.

The "problem" with Bayer is that the unavoidable colour aliases tend to be close - too close - to the wanted resolution. The only solution is to increase the photosite count (to move the colour aliases to higher frequencies) - which means then not taking full advantage of the chip resolution.

This is exactly what the C300 does. The coloured aliases become far higher frequency (expect to see them at about 2160 lpph) which is why they aren't seen on charts which only go up to 1200lpph, and why only luminance aliasing is seen on such - that's based on the 2x2 blocks at far lower frequency.

Which neatly comes back to the total absence of coloured aliasing on both the FS100 and the FS700 charts. There's no mystery with the FS700 - everybody agrees it has a higher photosite count, so the argument is identical to the C300. As for the FS100, accept it also has a high count sensor, and again the mystery goes away.

One possible way to knock this on the head once and for all is to look at a chart with res of around 2000-2400. If the FS100 shows coloured aliasing around there, that proves 100% it must have a high count sensor. I previously said "If you get the chance, try doing them with camera framed such that the chart only occupies half the width and height of the frame, which will effectively give detail up to 1600lpph." On second thoughts, if anybody does get the chance, try it with the chart only occupying a third the width, to give up to 2400pph.

[Alister Chapman]

Please take a look at the attached image. 3 Zone plates, Bayer (XF105), F3 and MC1P (clearvid).

I realise that the XF105 sensor is lower resolution than the F3, but you can clearly see the reduced diagonal resolution of the bayer sensor a trait not matched by either the F3 or MC1P. However the coloured aliases of both the bayer and clearvid sensors are on the same axis and this is because the R and B samples on a clearvid sensor are arranged on the H and V axis the same as bayer, it is the green photo sites that are arranged diagonally which results in the improved diagonal resolution.

Most of the zone plates I have seen for the FS100 do exhibit coloured aliasing all be it at a lower level than the F3, but the nulls are in the same places as the F3, thus a similar structure and pixel count may be assumed.

Of course, the F3/FS100/FS700 may not be bayer and they may not be clearvid, maybe something else. Interestingly as a side thought, if the FS cameras have a tilted array it may explain the strange flickering overload effect that you can get as pan across strong verticals and horizontals as the number of green pixels sampling V an H increases and decreases on alternate lines.

[Michael Stewart]

Hi David, you wrote

[/QUOTE]What strikes me is how similar the charts are between the FS100 and the FS700 - and how different the F3 chart is to both of them. This is especially the case when the hor/vert resolution imbalances are considered for the 100/700, I can't believe it's just a coincidence they are so similar in each case? Taken together with the facts that the FS100 and 700 come from the same factory and have so much else in common, whilst the F3 comes from a different factory and is so different in so many ways (not least being a widely different power consumption from either of the other two) I now think there's little doubt that the F3 has the 3.3 million sensor and the FS100 shares the 11.6 million sensor of the FS700.
[/QUOTE]

My understanding is because the chip is a 4K chip it has to do some electronic work to convert down to the 1080P (discard info?) which could cause it to look closer to the FS100. just a thought.

Mike

[David Heath]

Originally Posted by Michael Stewart

Hi David, you wrote

............

My understanding is because the chip is a 4K chip it has to do some electronic work to convert down to the 1080P (discard info?) which could cause it to look closer to the FS100. just a thought.

The simplest thing for the FS700 to do would be the same as the C300 (also with a 4k chip) - directly read out one "output pixel" from each 2x2 Bayer block. It's simple to do, requires minimal processing compared to normal deBayering (and hence low power) and from a 3840x2160 chip will directly give good 4:4:4 full 1080 resolution. (See ProVideo Coalition.com: Camera Log by Adam Wilt | Founder | Pro Cameras, HDV Camera, HD Camera, Sony, Panasonic, JVC, RED, Video Camera Reviews if you're not familiar with the C300 theory.)

The big question is why the horizontal resolution is so much worse than vertical? Why are the zone plates not symmetrical? (As they are with the F3.) And the real point that is that in many key features the FS100 and FS700 charts are so similar (same asymmetry, same slight aliases at 700 lines horizontal res, same lack of coloured aliasing etc) - and both very different to the F3.

This is why I consider the FS100/700 to share the same chip and be doing very similar "electronic work" to derive 1080 from it.

Th alternative explanation is that the FS100 has the 3.3 million chip of the F3, but with different processing ends up with highly similar characteristics to the 700!?! That seems far less likely than the other explanation - especially as it's the FS100 and 700 that come from the same factory.

[Lee Mullen]

Originally Posted by Matt Davis

Got a chance to play with an FS700 very close to final release specification. Currently running tests on all sorts of things, but I'm no Alister Chapman or Adam Wilt, so my apologies in advance.

However, keen to share this early finding.

Zone plate shot on FS700 and FS100 - same lens, same setup. Tried 720p mode as well as 1080p mode on FS700. I'm impressed. Yes, there's a little bit more aliasing, but that's the price you pay when the detail's way up on the FS100. The forum previews look horrific, so you'll have to pixel-peep the full frame downloads. I checked them against the PNG originals, and you're not missing much.

What on earth are they sir??