In this article (sorry its german) I read about camera sensors. They show values like 1,4 µm Pixel, or 1,55 Pixel. What does this mean, and how is it related to sensor size? If I got it right, a bigger sensor is better. For example this sensor has 1/2.55" in size, which is smaller than for example the IMX260 with size 1/2.5". But what about this measure of pixels, and what does it says about image quality?
When comparing camera sensors, what does [x] µm Pixel mean, and what does it say about image quality
It is the size of one pixel. 1 µm (micrometer) is a 1/1000 part of millimeter (there are 1000 micrometers in one millimeter, or 1000000 micrometers in a meter).
There is no (direct) relation to sensor size. But knowing sensor size and pixel size you can (roughly) calculate resolution.
The smaller pixels are, the more details are (theoretically) possible on photos. Unfortunately, smaller pixels lead to more noise on photos.
Okay, thank you. So in a sense its a trade-off between detail and noise.– StefanHFeb 20, 2017 at 13:13
2@StefanH No, I wouldn't say it's a trade-off. Lenses limit the quality of images much more than pixel size and noise level of modern sensors is almost negligible in usual everyday cases. ISO value has more influence on noise level than pixel size.– ZenitFeb 20, 2017 at 13:39
more noise-> yes, but as a side effect; it's worth understanding the root cause, which is that a smaller pixel simply captures less light. A camera will perform better in low light with larger pixels, requiring lower ISO sensitivity, leading to less noise. (In theory, a very powerful night camera would have a gargantuan sensor with large pixels). Dec 15, 2019 at 4:22
If I got it right, a bigger sensor is better.
Not necessarily. Depends on what you plan shooting.
When taking photographs of very distant objects such as moon or birds, slightly smaller pixels than commonly present on full frame sensors might be beneficical. Then you can get long telephoto reach with smaller lenses. For example, full frame commonly has 6 µm x 6 µm pixels with 24 megapixel resolution, whereas crop sensor could very well have 3.75 µm x 3.75 µm pixels.
To achieve 3.75 µm x 3.75 µm pixels, you can buy a full frame camera with 61.44 megapixel resolution, or a crop sensor camera with 24 megapixel resolution. The latter is cheaper, and using the former with long telephoto reach means you just crop the final image, making most of the 61.44 million pixels unused.
But anyway, there is a limit to how small it makes sense to shrink the pixels.
Here is a picture of moon with 400mm focal length on full frame:
Here is a picture of moon with a small sensor camera, CoolPix P1000, that zooms up to 3000mm equivalent (source):
Which do you think is better?
The ability of the P1000 to zoom to 3000mm equivalent is created by making the sensor very small, thus making the pixels very small. Unfortunately, they made the pixels so small that the small pixel size is starting to reduce resolution due to effects such as diffraction.
So, my main point by comparing these moon shots is to demonstrate that even though smaller sensor can easier have longer telephoto reach, there's a limit to how small you should make the pixels.
They show values like 1,4 µm Pixel, or 1,55 Pixel.
All of these values are way too small. They are not optimized for image quality but rather optimized for low sensor production costs and small camera size.
Anything below 3 µm or so is lens limited in practice. For example, 24 megapixel full frame camera takes better pictures than 24 megapixel crop sensor camera, the reason being that the pixels on the full frame camera are larger and don't magnify lens limitations as much.
Pixel size is just the area of each pixel in the sensor. It has no direct relation to sensor size, but the more pixels you pack in a sensor size, the smaller they have to be to fit into the sensor.
In general terms there is a correlation between sensor size and some aspects of image quality like noise and dynamic range. There are exceptions, but in general the bigger the sensor (all other factors equal) the less noise the image has.
One very common misconception is that pixel size has direct effect on image noise. As Tony Nortrhup explains on this video https://www.youtube.com/watch?v=_KYvp8PrCFc&t=1s there is no correlation between pixel size and noise because if the sensor is of the same size, smaller pixels are compensated by a greater number of them.
I received a negative vote for this response but I don't know why. Is there anything wrong with it?– MarcosOct 30, 2017 at 3:52
to summarize . smaller the µm size in a large sensor produces more compressed pixels and as a result it produces sharp good quality images. Noise on the other hand is on the software processing side. A good image processing software results to a good image quality reducing the unwanted noise. Also We can determine a good sensor on its raw shot taken(raw shots will take up more file size than usual jpeg images.
This answer is not very clear, and seems to not separate concepts very well. Dec 15, 2019 at 12:05
There are physical limits to how small you can make the pixels.
If you end up with pixels smaller than the wavelength of the light you are using photographically ( 0.3-0.8um, so around 0.5um2 ), you will not gain useful resolution. Similar limits apply to optical microscopes. Also, added optical structures (color filters, microlenses) over the sensor will likely misbehave in unexpected ways.
Also, the absolute possible resolution of lenses is limited by diffraction, no matter how you scale their size down. An f/2 lens has is diffraction limit for green (relative short wavelength!) light stated at ca. 1000lp/mm, will never do more than 2000 pixels in a row per mm - less for red light. This would also limit the useful pixel smallness to around 0.5um2.
Also, since sensor pixels are in the end photon counters, the smaller you make them, the steeper the "steps" in the count become - in the most pathological case, in very low light, it is up to random chance whether two or one or zero photons will hit the pixel - which would practically turn the output into white noise. Increase the area, and things will average out much smoother.