Image feature extraction

I’m looking for a way to take a source image end up with x/y coordinates of significant or important pixels from the image. I think this is similar to open cv style feature extraction.

Ideally this would either use the open cv imagepack nodes, or a dx11 compute shader.

any ideas?