How can I do a convex smoothing to a set of points in Matlab?

Hi

I'm using the function spline to do a cubic spline interpolation, nevertheless given the nature of that functions some parts are not convex. I need to ensure the convexity of the smoothed function, is there a way to do something similar like the spline but keeping it convex?

Many Thanks

[Image Analyst]

Please post a screenshot (type alt-printscreen) to http://snag.gy so we can see what you're talking about.

Answer 1

[Matt J]

This FEX file let's you constrain a spline fit to be convex

http://www.mathworks.com/matlabcentral/fileexchange/24443-slm-shape-language-modeling

[Image Analyst]

What does convex mean in this context? If your widely sampled "curve" is a sawtooth waveform (say it's peaks and valleys of a sine wave), what would a more finely sampled, interpolated, convex spline interpolation look like?

[Matt J]

I haven't used SLM myself, only noticed that John had a convexity-enforcing setting there. I would assume it forces the interpolated spline to be convex. So, in your example with the sawtooth, I assume the resulting fitted curve would just be flat at zero.

[Image Analyst]

Well then it wouldn't go through the knots. It would be a regression rather than an interpolation. Or do some kinds of splines not need to go through the knots?

[Matt J]

Yes, it would be a regression, but I assume that's what the OP wants, as the title of the post mentions "convex smoothing".