SKIP to Questions below if you want :

In super resolution I can create a high resolution signal from a low resolution signal. But it is also possible that the low resolution signal contains aliasing. The problem with aliasing is that it cannot be distinguished between the real low resolution frequencies. But in super resolution you can make multiple low resolution measurements with aliasing and calculate the shift between the signals (afaik) that can help to reconstruct a higher resolution signal. Check some papers like here : Multi-Frame super resolution, SR from aliased images

What I am thinking is if I can make multiple shots from one scene with aliasing and construct a higher resolution image out of it, how could I apply this to vibration from a machine. In the first paper google shows how they use the vibrations from a hand for taking multiple photos from one scene. The vibrations from a human hand help to capture different frequencies with aliasing. Later on, all merged can be used for super-resolution.


  1. I have a fixed sensor on a machine which basically means I have a camera fixed on a stand(compare photo camera wich takes multiple pictures at same time from the same scene with aliasing to construct a higher resolution). How would this affect the super-resolution in comparison to using different angles in parallel to have multiple measurements with different offsets?
  2. Could I apply this on vibration with making a long measurement, dividing the measurement in different blocks and trying to merge them as they where different captures, instead of using different angles?

I don’t think that you can do multi-frame super resolution in the traditional sense if there is no camera rel scene movement, nor if the camera applies «proper» Nyquistian spatial prefiltering?

I think of good old interlacing as an example of potential super resolution. You get a 2-frame cadence consisting of separate time and spatial samples, usually moderately filtered. If there is no movement, you can get up to double spatial resolution. If there is movement that can not be resolved, you get high temporal resolution.

  • $\begingroup$ I don't understand your answer or is it a question :D? what do you mean by camera rel scene movement? Maybe i have formulated the question wrong or stupid.. but i want to know if i have a fixed sensor on a machine and make a long measurement with aliasing .. can i build blocks of the measurements and combine them together to get a higher resolution`? Thats how it is done in multi-frame superresolution. The difference would be that there is no "movement" like it is in a camera frame superresolution $\endgroup$ – Khan Mar 29 '20 at 20:14
  • $\begingroup$ If the camera does not move relative to the scene, then I don’t see how it could be done, as each new frame would not add any new information. Such movement could be camera movement, sensor movement, or objects moving in the scene. $\endgroup$ – Knut Inge Mar 30 '20 at 3:07
  • $\begingroup$ I thought its not about changing the scene, but having different high-frequency artifacts with aliasing. Applied on a machine with vibration that would mean, the 'scene' is the vibrating machine and if the vibration changes a little over time, I can make different measurements with high-frequency components. The other thing would be to put 3 vibration sensor on a machine and you have automatically a different 'scene' because the same vibration appears different from different angles $\endgroup$ – Khan Mar 30 '20 at 6:56
  • 1
    $\begingroup$ Perhaps I did not understand your question. Yes if you are vibrating the camera, you have the possibility of aliasing. If the exposure duration is as long as the time between each frame, you will have motion blur that is acting to remove aliasing. This can perhaps be solved by having shorter exposure times. $\endgroup$ – Knut Inge Mar 30 '20 at 13:51
  • $\begingroup$ I am not using a camera at all. It was just an example. My question is regarding a vibration sensor which has three axes, x,y,z. And if it is possible to apply super-resolution on this sensor. And if it is possible to apply 'multi-frame-sr' on the axis because all axes do measure simultaneously (e.g each axis is a frame(?)) $\endgroup$ – Khan Apr 6 '20 at 7:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.