Pixel aspect ratios

[icypix]

Folks -

I'm hoping someone on this forum can point me in the right direction on this. I currently use a Nikon D-80 and D-90 in my lab. A colleague is using a Sony XC-73CE in his lab for similar work. We are attempting to compare our measurements which involve the determination of microscopic areas. Our individual set ups provide us with scale factors in pixels per cm on each digital image. Our problem is that we cannot get from our linear calibration of px/cm to px^2 per cm^2 without knowing the aspect ratio of the pixels in our respective cameras and I do not wish to assume that they are square (1:1). Two weeks ago I submitted this question to Nikon for my cameras and they have not responded. There is no information on this in the camera manual. Additional inquiries to Nikon produce no response. Would anyone here have a suggestion as to how to proceed?

Many thanks!

[GrahamS]

You don't say where you are located or which company you work for, but I suggest that you try to establish communication with Nikon's Instrumentation division in Japan, directly. Try here: https://www.nikon-instruments.jp/eng...act.aspx?KBN=4

Alternatively, contact the local Nikon office by 'phone, not eMail. Ask for the local rep to call on you.

[NorthernFocus]

I think you are overthinking this one. You may be confusing pixel with photosites on the image sensor. The pixels in an image are a computer generated interpretation(bayer interpolation etc.) of the light falling on the image sensor. So it is safe to treat the pixels as square. After all if they weren't square the the 1x1.5 aspect ratio of the sensor would not reflect in the same ratio of pixels.

Then again, maybe I'm oversimplifying.

[benm]

The D80 has 3872 pixels horizontal x 2592 pixels vertical. The horizontal dimension is 23.6 mm and the vertical dimension is 15.8 mm. 3872/23.6 and 2592/15.8 are both 164 pixels/mm, leading me to believe they are square. Similarly for the D90 - the pixels are 4288 x 2848 and the dimensions are the same as the D80 (works out to 181).

All figures obtained from DPreview's website.

[xpatUSA]

Folks -

I'm hoping someone on this forum can point me in the right direction on this. I currently use a Nikon D-80 and D-90 in my lab. A colleague is using a Sony XC-73CE in his lab for similar work. We are attempting to compare our measurements which involve the determination of microscopic areas. Our individual set ups provide us with scale factors in pixels per cm on each digital image. Our problem is that we cannot get from our linear calibration of px/cm to px^2 per cm^2 without knowing the aspect ratio of the pixels in our respective cameras and I do not wish to assume that they are square (1:1). Two weeks ago I submitted this question to Nikon for my cameras and they have not responded. There is no information on this in the camera manual. Additional inquiries to Nikon produce no response. Would anyone here have a suggestion as to how to proceed?

Many thanks!

I can tell you that the D90 pixel pitch is different between horizontal and vertical, from the sensor size and resolution:

Horizontal: 23.6mm / 4288 = 5.504 um
Vertical: 15.8 / 2848 = 5.548 um

The wolves might be circling to say I've only used the active pixels, but it's close enough, trusssst me.

The aspect ratio of the effective pixel area will not help in figuring out your calibration. Use the pitch, I reckon. [edit] . . .

But, thinking aloud, is it that easy? Say an object is photographed at 1:1 with a macro lens and say the object is irregularly shaped so the lab guy uses, for example, ImageJ to count the pixels in the shape. Or, maybe a histogram after deleting the background somehow.

I guess you have to assume an area implicitly, i.e. (23.6x15.8)/(4288x2848)=a D90 pixel 'pitch area' of 30.53334um^2. So for e.g. a count of 10,000 pixels the object area at 1:1 magnification is 305,333um^2. <-- oops, edited slightly, duh

Good Luck!

[revi]

"icypix" (for lack of a better name :
The best way would be to do a calibration in both directions to determine aspect ratio. You only need to do that once
for each camera, and it's better (and probably faster ) than discussing from published sensor data...

Could you try and get another measure for the pixels/cm by turning the camera from landscape to portrait orientation?
Alternatively, can you get a good picture of a perfectly circular object?
Or would you have access to a target/object with precise horizontal and vertical dimensions?
Either of those would help answering your question directly, w/o discussions about pixel pitch, visible vs. total area etc.

The tricky bit with calculating pixel pitch is the precision of the sizes, which are rounded...
In Ted's example, his values are valid for those exact measurements, using normal admitted practice of assuming an error
of +/- 0.5 in the last digit given, I get pitches of
Hor: 5.49 -- 5.52 µm (23.55 / 4288 = 5.492; 23.65 / 4288 = 5.515)
Vert: 5.53 -- 5.57 µm (15.75 / 2848 = 5.530; 15.85 / 2848 = 5.565)

And that's assuming that the values given by the manufacturer are correct, and we can ignore the 'inactive' pixels.
And I'm not sure about that: the sizes indicate an aspect ratio of 1.4936 (1.486--1.502), pixel counts give 1.506.
Assuming square pixels, that means that there's only ~30 pixels 'too many' in the horizontal direction.
And looking at the values that give an aspect ratio of 1.502, pixel pitches are very close: hor: 5.515 µm & vert: 5.530 µm.

Another thing: are the deviations from square relevant to OP's case?
From Ted's data, we get to differences in pixel pitch that are at worst 80 nm, or <2%, and at best 15 nm (or < 0.3%).
If that's relevant in OP's case, only he can tell us.

[dje]

As the others have indicated, calculations based on sensor size and pixel numbers indicate that the pixels are approximately square. I've only ever heard of non-square pixels in the world of digital video.

A couple of other things to consider

EXIF data contains info on x and y resolution but I'm not sure how useful that is. From what I can see, the resolution is usually the same in x and y directions.

Photoshop seems to open up an image with the par assumed to be 1. This can be changed but all that does is change a flag in the file and is used mainly I think where stills are to be imported into video.

I guess it depends on the accuracy you are looking for but I would have thought you'd be fairly safe in assuming square pixels. I don't like your chances of getting anything out of Nikon on a question like this - after all, these cameras are not made as scientific instruments.

Dave

[icypix]

Many thanks to all. I have just heard back from Nikon that the pixels for both cameras are indeed square.

[dje]

Many thanks to all. I have just heard back from Nikon that the pixels for both cameras are indeed square.

Glad to hear you got an answer and thanks for the update.

Dave