1
$\begingroup$

My master’s dissertation subject is to analyze and implement an algorithm that would sample (store) thermal images from camera with different frequency rates to optimize database size and processing power.

The thermal images are provided to computer with constant flow with packets structure consisting of header and matrix of 14 bits values per each ,,pixel''.

From publications that I found, most commonly compression techniques are used, but here I should develop dynamic sampling algorithm without althering data structrure.

Is anyone aware of any publication related to this subject? It seems that I cannot find any information on sampling patterns for raw 2d data streams, nor algorithms that would detect or calculate if the image differs enough from previous ones and is worth capturing.

$\endgroup$
6
  • $\begingroup$ Well, we don't have any information about the structure of your data, and anything but entropy coding (e.g. huffman) will need to have such knowledge. Chances are it's pretty similar to usual video, so have you looked into video compression? $\endgroup$ Apr 7 at 11:36
  • $\begingroup$ @MarcusMüller the thing is that it can't be compressed but rather sampled in different rate. $\endgroup$
    – Andre
    Apr 7 at 11:58
  • $\begingroup$ Example usage ... when cooled object starts to gain temperature the overall slope is high so given that camera can provide 60 images per second all should be stored. But after some time the diffrence is less so images should be stored less often. I think thats the general conception of the subject of disertation. $\endgroup$
    – Andre
    Apr 7 at 12:04
  • $\begingroup$ well, sounds like video compression, a lot. They do simply not store things that don't change, and there's interpolation between what they call keyframes, which leads exactly to what you want (non-constant sampling times for individual values). I think you should probably get a solid base in video encoding :) $\endgroup$ Apr 7 at 12:16
  • $\begingroup$ In absence of that: try per-pixel linear precoding, this would yield the moments you do need to update. $\endgroup$ Apr 7 at 12:54

Your Answer

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

Browse other questions tagged or ask your own question.