Upcoming hands-on comparison of F3, FS100 and AF100 from Philip Bloom

[Alister Chapman]

A few important notes Brian.

No one outside of Sony truly knows how the pixels on the F3/FS100 sensor are arranged. We have been told Bayer, but that may mean quite a few things. Normally bayer is a pattern of groups of 4 pixels arranged like this:

R G
G B

Normally from this arrangement you can accurately extrapolate useable resolutions around 70 to 80% of the horizontal pixel count, which for the F3 sensor is calculated to be 2456x1372, this gives a ball park theoretical resolution of around 1840 pixels or around 920 LW/PH. Nyquist for 1920x1080 is 960 LW/PH, Any resolution beyond this will contain aliasing aterfacts which are undesireable, so a perfect camera would have a resolution of 960 LW/PH, but there are many issues which mean that we cannot create the perfect camera so compromises will be made.

My MTF50 results are not the finite resolution of the cameras as above.

MTF50 is where the contrast between a black and a white line becomes 50%, or mid grey. This is important because this is the point where the viewer normally stops seeing any fine details as the contrast between the details is so low that they cannot be perceived, even through they may still be there. It is quite normal to have the finite resolution around 25-35 percent higher than the MTF50, but this will vary from camera to camera, system to system. MTF is also a measure of the system, i.e. camera plus lens plus processing. As a guide if you see an MTF50 figure of 750 LW/PH then the finite resolution may well be around 900 - 1000 LW/PH which does tie in quite well with what has been measured from the F3 and FS100 and the pixel numbers supplied by Sony. In addition it's not a bad result.

MTF50 also tells you about contrast in the mid range and how sharp the image will look with less fine image details. This is important as even with a substandard display this difference will still be seen.

So both the F3 and FS100 are definitely at the good end of 1920 x1080 camcorders in terms of resolution. The F3 has a higher MTF50 than the FS100 and this will likely lead to slightly better mid range contrast and perceived sharpness.

Sorry Steve no VG10 resolution tests. I'm prepping for a big shoot at the moment and the PC is in the my other office, not sure I will get a chance to do it this week.

[Chris Barcellos]

Allister;

I have been following this debate from the beginning, not understanding half or more of the technical arguments. Thanks for posting this last explanation. Really interesting crystalization of what you guys are talking about in terms of technical aspects of our mysterious machines.

[Brian Drysdale]

Thanks for the notes Alistair

I can see the FS 100 being used on larger productions as a B camera, where space is tight or its light weight could be an advantage for rigging a camera,

[Alister Chapman]

Steve, David: Don't know if you know the definitive answer to this, but I have been looking more at interpreting the very subtle clues that zone plates give about sensors, in particular bayer. One thing I had thought is that when you see red/blue colored rings i.e. colored aliases (not to be confused with the rainbow green, red, blue moire rings caused by lack of luma resolution) on the horizontal and vertical axis that are typically out of step with the luma aliases it is a measure of the difference in resolution between the G and the R/B samples typical of a bayer sensor. Can anyone confirm this or have I got this wrong?

Looking at both mine and Alan Roberts zone plates that explore beyond Nyquist you can see these strongly coloured aliases, to me this suggests a conventional bayer sensor, assuming that I have the above correct.

Looking at a Q67 sensor with the bayer matrix basically rotated through 45 degrees these blue/red aliases shift through 45 degrees and now appear on the diagonal axis, so that rules out Q67 for the F3/FS100.

[David Heath]

Hello Alister,

Looking at Alans zone plate ( http://thebrownings.name/WHP034/pdf/...ony_PMW-F3.pdf ) then what I find most revealing is what you can see in figure 2 - the centre monochrome pattern, not the big red/green zone plates. That gives the out of band response, and here we basically see the lower examples of eight alias centres, all in a square, with four at the the corners, four in the centres of the sides. The corner ones are green/magenta, the vert/hor ones are more orange/blue.

It's my belief that the dimensions of this square is fundamental with the sensor dimensions, and backs up the figures given by Sony as being of the order of 3.5 million photosites.

Compare it with the diagrams here - ProVideo Coalition.com: Camera Log by Adam Wilt | Founder | Pro Cameras, HDV Camera, HD Camera, Sony, Panasonic, JVC, RED, Video Camera Reviews and for the BAYER sensor note that the colours of the alias rings are exactly as for the zone plates for Alans F3 tests. This suggests very strongly that the sensor not only does have the photosite count stated by Sony, but is also (as they say) Bayer, not Q67.

As I understand it, the beauty of Q67 should be that it doesn't offer a magical resolution increase over Bayer for a given photosite count - but for a video usage should mean high quality with far easier processing. It relies on being implemented with twice as many photosites as system resolution - so 2x1920x1080 photosites for a 1920x1080 system. See ProVideo Coalition.com: SONY Livewire | Vendor Chanel - each output pixel (bounded by white line) is formed from one unique green photosite, and two blue quarters and two red quarters. That seems far easier than Bayer processing, and will mean a direct 1080 output - no need for downconversion.

I also think it's important to realise that the subtle clues from zone plates aren't just generated by the sensor dimensions and layout - but by the way the photosites are read and processed. This becomes particularly true for DSLR video - zone plate patterns from something like a 5D are far from what the basic sensor layout would suggest!

[Steve Mullen]

Originally Posted by Alister Chapman

Looking at a Q67 sensor with the bayer matrix basically rotated through 45 degrees these blue/red aliases shift through 45 degrees and now appear on the diagonal axis, so that rules out Q67 for the F3/FS100.

I too doubt the use of Q67 or Striped Filter.

But, if the sensor has about 4 million photosites and it is Bayer -- then what are the "specifications" that Juan says he cannot reveal?

One clue from Roberts' is that he notes is the MISSING extra strength of the Green signal which he says traditionally comes from the Bayer having twice the number of green photosites as either Red or Blue. This is why I assumed 8x8 Bayer photosites (with the extra Green sample) were combined into RGB pixels where all three signals had the same strength. The combining was, I assumed, also the way sensor sensitivity was increased.


There is a version of the Bayer filter that replaces one of the extra green filters with a translucent filter. This type of filter would show equal RGB signal strength AND, of course, the point of the translucent photosite is to increase sensitivity.

"The panchromatic pixels have no filter so they are at least a stop faster than the color pixels. Half the pixels are panchromatic so they provide the backbone of the image (a stand-in for the luma channel). First, interpolate the missing panchromatic values at colored pixelsites. Second, the RGB signals are treated like chrominance channels: R-P, G-P, B-P. Once they are interpolated and cleaned, they are added back to the panchromatic values and the result is hi-res RGB. There is a combined hardware / software paradigm shift. The panchromatic pixels provide the photographic speed and record the scene detail. Once the colors have been interpolated using the panchromatic values as a guide, the panchromatic record is cancelled out and the colors remain and the color errors are no worse than for the Bayer pattern."

Perhaps the sensor's extra sensitivity is a function of both the 4X larger photosites AND the translucent photosites.