Where the PDAF sensors are...

SHood wrote:

Thanks Helen. I had wondered how the Nikon 1 sensor was doing PDAF. I am surprised no review site has picked up on this or at least asked Nikon how it was done.

I was just having a look around and there's a sort-of mention of this in the DCResource review of the J1 - well, it links to a Hungarian review where they got a line showing a bit when it happened to run at just the right (or wrong!) angle to a horizontal line on a test chart they were shooting. I think this was just a single line that showed in part, whereas I accidentally stumbled upon a way of showing the lot of them.

Nikon 1 state to have 73 autofocus point with the possibility to "choose any of 73 focus points when using the single-point AF mode "

Because there's so many I can guess they are covering about the whole sensor and if anything are more dense near the center as well as the 1/3 rules lines. I also doesn't make sens to allow to choose between points that are at the extreme of frame like that.
http://imaging.nikon.com/lineup/acil/bodies/v1/features02.htm

on another note, if you have a Nikon one... you can probably set it to single AF and loop thru all the point site.

SHood wrote:

Helen, I assume that this image was a crop and those lines represent a 1 pixel height. Could you confirm?

It's not a crop, as it happens. It's the whole frame, so we've got quite a large part of it "covered" by PDAF sensitivity. Obviously the image quality in this shot is a bit rubbish due to it being a moving TV image - it's a 37 inch screen shot from about 9 feet away with the 30-110mm lens at (checks EXIF) 57.2mm (154mm equivalent in film terms), 1/40 sec. It is a crop of the TV screen image, because at that zoom level, it more than fills the frame of the shot. But the shot itself isn't cropped.

In case anybody wonders if it's some property of the TV screen that's being shown, I'm pretty sure that it isn't, since the lines occupy the EXACT same pixels on each shot I took, even though I wasn't framing up the TV with any sort of precision (and being a rather slow shutter speed for the focal length, the VR was of course moving the image occasionally, too).

SHood wrote:

In that case these lines are "thick". Can you tell from the original image how many pixels in height are these lines?

Well, it's really hard to be sure, but it looks like a maximum of 3 pixels high. In parts though, only 1 pixel height appears discoloured. So the pixel rows above and below might just be some sort of interpolation, like how a single stuck pixel ends up with an array of affected pixels around it forming a larger cross pattern, due to the Bayer interpolation pattern.

Helen wrote:

SHood wrote:

Helen, I assume that this image was a crop and those lines represent a 1 pixel height. Could you confirm?

It's not a crop, as it happens. It's the whole frame, so we've got quite a large part of it "covered" by PDAF sensitivity. Obviously the image quality in this shot is a bit rubbish due to it being a moving TV image - it's a 37 inch screen shot from about 9 feet away with the 30-110mm lens at (checks EXIF) 57.2mm (154mm equivalent in film terms), 1/40 sec. It is a crop of the TV screen image, because at that zoom level, it more than fills the frame of the shot. But the shot itself isn't cropped.

In case anybody wonders if it's some property of the TV screen that's being shown, I'm pretty sure that it isn't, since the lines occupy the EXACT same pixels on each shot I took, even though I wasn't framing up the TV with any sort of precision (and being a rather slow shutter speed for the focal length, the VR was of course moving the image occasionally, too).

Pho3NiX wrote:

Nikon 1 state to have 73 autofocus point with the possibility to "choose any of 73 focus points when using the single-point AF mode "

Because there's so many I can guess they are covering about the whole sensor and if anything are more dense near the center as well as the 1/3 rules lines. I also doesn't make sens to allow to choose between points that are at the extreme of frame like that.
http://imaging.nikon.com/lineup/acil/bodies/v1/features02.htm

on another note, if you have a Nikon one... you can probably set it to single AF and loop thru all the point site.

Yes, you can. I read that page - it's interesting. So, 73 (maximum) AF targets in PDAF. The manual only mentions that there are 135 total focus areas in single-point AF, and 41 in auto-area. These totals will include the CDAF points no doubt, which are smaller (I suspect) and inherently more "mobile". The figures are the same for the V1 and J1, by the way. The cameras seamlessly switch between the two systems according to the light and contrast levels, so it's difficult to be certain about how it acts - except that the PDAF variety is pretty much instant and clearly in use in bright light.

Helen wrote:
. . .

Helen, I assume that this image was a crop and those lines represent a 1 pixel height. Could you confirm?

It's not a crop, as it happens. It's the whole frame, so we've got quite a large part of it "covered" by PDAF sensitivity. Obviously the image quality in this shot is a bit rubbish due to it being a moving TV image - it's a 37 inch screen shot from about 9 feet away with the 30-110mm lens at (checks EXIF) 57.2mm (154mm equivalent in film terms), 1/40 sec. It is a crop of the TV screen image, because at that zoom level, it more than fills the frame of the shot. But the shot itself isn't cropped.

In case anybody wonders if it's some property of the TV screen that's being shown, I'm pretty sure that it isn't, since the lines occupy the EXACT same pixels on each shot I took, even though I wasn't framing up the TV with any sort of precision (and being a rather slow shutter speed for the focal length, the VR was of course moving the image occasionally, too).

Helen, if you edit your DPR profile, there's a setting at the bottom of the "Privacy and safety" tab that will allow us to view and download your original, full size photos for closer examination.

micksh6 wrote:

This may be an effect of PDAF sensors but there can be some other explanation.

Yes, I initially thought that.

It could be a result of some weird interference between LCD refresh rate and exposure time and aliasing (pixelation) could make it more pronounced.

And I suppose it still could, even if it does result in in showing the position of the PDAF arrays - after all, something about the images of my TV screen causes the camera to fail to hide the lines as thoroughly as usual.

Is your TV 60Hz or 120Hz? 1/40s exposure time could be problematic - it may expose interference artifacts. And incandescent light too - it is also 60Hz in US, as well as LCD backlight refresh rate. LED vs CCFL backlight may matter here too.

It's neither - it's 50Hz or 100Hz, because I am in the UK and that's our rate (electricity supply is 50Hz and it's the PAL TV system). The reason I'm not sure of the figure is that the specs of the TV are a bit hard to decipher, due to marketing-speak. There is a close equivalent TV on the US market, though, I believe. My TV is a Samsung UE37D6530; an LED backlit model from this year's UK range. I believe the panel in it is a 100Hz one though when the TV's accepting input from my PVR (though this picture and others I will upload was straight from the TV's tuner) it labels it as 50Hz. The set has the motion processing (CMR) set to "Clear" and the flickering backlight is not enabled - I think if it was, that would fulfil the marketing-speak that allows them to claim it can act at the equivalent of 200Hz (mentioned on the packaging!).

Did you try different exposure? Would be interesting to see 1/25, 1/50, 1/100 and 1/200, for example.

I've examples taken at 1/20, 1/25 and 1/60 as well, which I shall upload in due course. Unfortunately I've got to do some other family duties before I can get onto that.

I understand these lines are always on the same position on the sensor, right?

I think so, unless I've looked at them too quickly and jumped to conclusions.

Can you look at RAW image in software that doesn't apply distortion correction? I don't know such software for Nikon 1, perhaps Picasa or RawTherapee? And if you shoot LCD when camera is rotated?

So far I've only shot them as jpgs. I'm not a big fan of RAW, so the only thing I have to hand I think is the latest version of ACR - don't know whether that applies distortion correction and I've only just had the software it's in (trial version!). I'll try and shoot some in RAW later and have a look.

It these lines belong to sensor it should be possible to find another subject that would "confuse" image processing and would also reveal the lines. For example, if you make a photo of some another high resolution very contrast photo with lots of details the image processing should fail at least in some places. You know the location of the lines so you can inspect them more carefully.

And even if these belong to sensor it could be a sensor defect. I'm sure sensor specs allow several dead lines and software compensates for that. You might have found a case when it doesn't work well. Perhaps others can reproduce it.

I hope not! I will be interested (and relieved, now you've raised the ghost of a malfunction) if others find them in their own cameras. Since these Nikon 1 cameras do have PDAF detection integrated onto the imaging sensor itself (and I remember reading that the portions which carry out the function cannot participate in image capture), I am hopeful that this is what they are, especially given their coverage pattern.

PS. I don't have Nikon 1. I'm just curious of possible PDAF effect on mirrorless cameras.

Billx08 wrote:

Helen wrote:
. . .

Helen, I assume that this image was a crop and those lines represent a 1 pixel height. Could you confirm?

It's not a crop, as it happens. It's the whole frame, so we've got quite a large part of it "covered" by PDAF sensitivity. Obviously the image quality in this shot is a bit rubbish due to it being a moving TV image - it's a 37 inch screen shot from about 9 feet away with the 30-110mm lens at (checks EXIF) 57.2mm (154mm equivalent in film terms), 1/40 sec. It is a crop of the TV screen image, because at that zoom level, it more than fills the frame of the shot. But the shot itself isn't cropped.

In case anybody wonders if it's some property of the TV screen that's being shown, I'm pretty sure that it isn't, since the lines occupy the EXACT same pixels on each shot I took, even though I wasn't framing up the TV with any sort of precision (and being a rather slow shutter speed for the focal length, the VR was of course moving the image occasionally, too).

Helen, if you edit your DPR profile, there's a setting at the bottom of the "Privacy and safety" tab that will allow us to view and download your original, full size photos for closer examination.

Thanks, I'll have a go - though it may be a few hours till I can.