Are pixels square or rectangular?

[CVhobbyist]

What I mean is that a pixel could represent a larger angle vertically than horizontally, or vice versa. I've measued that on Samsung Siii the pixels are square, at least to within about 0.5% which is my measurement error. Can I assume that this is so on all smartphone cameras? I would need to know because of an app I'm developing.

Samsung politely explain that they don't know because is not a property which has been explicitly specified, but a consequence of other specifications. (At least, that's about field of view in general, maybe physics make them square by necessity?)

 

[runnah]

Square.

 

[CVhobbyist]

Is it general for all cameras? Is it exactly square or could it be 1% rectangular (which is my measurement limit)?
I read about the Matlab Camera Calibration Toolbox that it has a method for dealing with rectangular pixels, so it seems to exist. But maybe in scientific equipment and not for consumers.

 

[KmH]

In the past some DSLR cameras had both square and rectangular pixels.

I doubt you can safely assume all cell phone camera image sensors have 1:1 aspect ratio pixels.

 

[Luigi_xp]

Yes, they probably have 1:1 pixels. i don't know exactly, but i've never seen an modern (CMOS) imaging sensor without square pixels.
Explain what you want to do and we can help more.

 

[JerryLove]

Given that multiple pixels are used to get a single "point", and the arrangement and orientation of those pixels varies: I think the question may make assumptions which render it unanswerable.

Take a peek at this brochure for an example of what I mean. EXR CMOS TECHNOLOGY | FUJIFILM

 

[CVhobbyist]

I've discovered that one can query Android's Camera Parameter class for its field of view!

Camera.Parameters.getVerticalViewAngle()

Camera.Parameters.getHorizontalViewAngle()

According to that, my Samsung S3 has 49.3 and 63.0 degrees FoV respectively. That's a ratio of 1.278.
But the number of pixels in highest resolution is 2448 and 3264. That's a ratio of 4/3 = 1.333...
One angular degree is 49.7 pixels vertically, but 51.8 pixels horizontally.
That's a substantial difference of 4.3%!

Should I adjust for that, or can I safely assume that pixels are square anyway?

It has proven a bit difficult to practically measure physical objects and count pixels between blurry JPEG edges to find out with good enough precision to be sure. I ask here before I set out to perform such measurements more ambitiously if necessary (including C++ programming to access some sharper raw image format which Android unfortunately hides).

Explain what you want to do and we can help more.

I want to measure the relative sizes of objects with high precision. If their sizes change in terms of numbers of pixels depending on how they are oriented, then I need to adjust for that.

Given that multiple pixels are used to get a single "point", and the arrangement and orientation of those pixels varies: I think the question may make assumptions which render it unanswerable.

Take a peek at this brochure for an example of what I mean. EXR CMOS TECHNOLOGY | FUJIFILM

Interesting! And kind of bad news.