0
$\begingroup$

I am pretty new to DSP and have been looking at how image scaling is done in hardware. I and came across a 2D scaler core for an FPGA here. I have read up a bit about polyphase filters, but am confused about what the "16 phases" referred to on page 3 with the following diagram are.

enter image description here

I have not come across this elsewhere and was hoping for some insight as to what is going on here.

My best guess is that they limit the available interpolation points to 16 discreet points in $x$ and the same for $y$, and so any fractional scaling where a sample point falls between $2$ of these $1/16^{\rm th}$ sampling points, will just use the closest point to it?

Any insight would be much appreciated.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

My best guess is that they limit the available interpolation points to 16 discreet points in x and the same for y, and so any fractional scaling where a sample point falls between 2 of these 1/16th sampling points, will just use the closest point to it?

Yes, that's it. Probably on the assumption that within 1/16 of a pixel shift, and with the further interpolation on the output, that filter will be good enough.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Image processing and "plain old" DSP tend to speak somewhat different languages, so the terms can vary. However, they're giving you a general tool to implement something called "polyphase filtering" in two dimensions. If you really want to understand that, you need to get a good grounding on sampling theory and filtering theory. $\endgroup$
    – TimWescott
    Commented Nov 9, 2020 at 15:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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