That mostly depend on the size of the feature you want to preserve/eliminate and if these features show a dominant orientation. It's difficult to be more precise without you providing some example.

P.S.: The more "work" you ask people to do to answer you, the less responses you will get! That's one of the inescapable trade-off of life!

Thank you dear Bruno. Let us say "Skin Cancer" image?

Even saying "skin cancer" image says nothing without the resolution and the apparent lesion size in pixel. Also, skin lesion, color and textures are important, two things mathematical morphology is not very good at.

How exactly would you apply mathematical morphology to these images?

For shape analysis, the best way is to use multivariate directional models defined on matrix manifolds. For directional statistics see books by K. V. Mardia and Drydon.

Dear Dr. Muhammad Ali,

Could you please upload the material "multivariate directional models defined on matrix manifolds", for us.

Thanks