It's well-known that Gaussian convolutions are separable. So are convolutions of first-order and second-order Gaussian derivatives. The former can be made steerable by multiplying ∂G/∂x by cos([itex]\theta[/itex]), multiplying ∂G/∂y by sin([itex]\theta[/itex]), and subtracting the result (please correct me if my math is wrong). I'm wondering if it's possible to perform a steerable convolution of a second-order mixed Gaussian derivative, i.e. ∂(adsbygoogle = window.adsbygoogle || []).push({}); ^{2}G/∂x∂y. A non-steerable convolution of that form is definitely separable - first convolve the input by ∂G/∂x and then convolve the result by ∂G/∂y.

This is not a homework assignment. I'm researching computer vision and image processing in my spare time. Second-order mixed Gaussian derivatives are apparently useful in determining saddle points in images. Thanks in advance for your help!

**Physics Forums | Science Articles, Homework Help, Discussion**

Join Physics Forums Today!

The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

# Steerable Gaussian Mixed-Derivative Convolution - Separable?

Can you offer guidance or do you also need help?

Draft saved
Draft deleted

**Physics Forums | Science Articles, Homework Help, Discussion**