Can somebody explain to me why the Horn and Schunck algorithm (probably mostly the smoothness constraint) is robust to image sequences that are quantized rather coarsely in space and/or time. (Thought that is only the case when I use a multi-resolution approach) Also it is insensitive to additive noise and quantization of brightness levels. It calculates the optic flow with a Laplacian of the flow velocities so that the complete image should have an effect on each image gradient vector. But isn't this a reason that the algorithm is robust to these characteristics?

Please provide a detailed explanation. If i've made mistakes by explaining my problem please point it out, this method is still very difficult for me to understand in detail.

  • $\begingroup$ I am not completely sure that it is robust(please provide some reference), but if that is the case my first guess is it is robust to noise because it operates on windows (usually 4 by 4) of pixels, so if your noise has a zero mean distribution, because of addition and subtraction process, the whole process becomes immune of noise. $\endgroup$ – MimSaad Aug 2 '17 at 13:50
  • 1
    $\begingroup$ It is written in the Abstract of "Determining Optical Flow" by Horn and Schunck from August, 1981. $\endgroup$ – baxbear Aug 2 '17 at 13:51

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.