cubic spline interpolation is required.

[ManikandanVA]

RTL code for cubic spline interpolation is required. will be highly appreciated. I am working on images.

[ZipCPU]

Cubic splines are often considered a gold standard of interpolation. They shouldn't be. 1) They aren't shift invariant, 2) require infinite data for data streams (not images) and 3) they require a complex matrix solution.

Other, better, solutions exist without these problems.

1. A quartic exists which will get you performance at least as good, but without the difficult requirements above. The resulting interpolator will remain smooth, easily interpolating between points, etc. It's quite doable. The development falls along the lines of the quadratic development presented both in this repository and in this post.

2. If you really need a cubic spline interpolator, then a series of FIRs can be used to generate approximate cubic spline coefficients--again, without the required matrix solution.

Both of these approaches eliminate the need for the matrix solution, and get arguably as good as or better performance than cubic splines.

Is there a reason why you need cubic splines specifically? Or would another approach also work for you?

Dan

[ManikandanVA]

hi I am using cubic interpolation for steganography application. I need a fucntinal verilog code for working upon

…

On Thu, Nov 25, 2021 at 7:30 PM Dan Gisselquist ***@***.***> wrote: Cubic splines are often considered a gold standard of interpolation. They shouldn't be. 1) They aren't shift invariant, 2) require infinite data for data streams (not images) and 3) they require a complex matrix solution. Other, better, solutions exist without these problems. 1. A quartic exists which will get you performance at least as good, but without the difficult requirements above. The resulting interpolator will remain smooth, easily interpolating between points, etc. It's quite doable. The development falls along the lines of the quadratic development presented both in this repository and in this post <http://zipcpu.com/dsp/2018/01/16/interpolation-is-convolution.html>. 2. If you really need a cubic spline interpolator, then a series of FIRs can be used to generate approximate cubic spline coefficients--again, without the required matrix solution. Both of these approaches eliminate the need for the matrix solution, and get arguably as good as or better performance than cubic splines. Is there a reason why you need cubic splines specifically? Or would another approach also work for you? Dan — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AUZNN46L2YNBPKJJ6JPXVR3UNY6OPANCNFSM5IVKQTCQ> .

[ZipCPU]

Thank you for your recommendation, then.

This is an unfunded feature, but if/when I get a chance to implement such I will do so.

Dan