Comment by lynnharry

There's no way to do it traditionally. Your request would need the algorithm to understand the content of the image. Deep learning based image vectorization probably has a similar objective.