Speed-up watershed segmentation using information from a binary image

[VPetukhov]

Following the example from the documentation, currently segmentation is performed on the whole image, and later it's filtered with the binary mask. Though, using mask during the segmentation would allow to significantly reduce computation time, avoiding computation for background pixels.

How about adding new optional parameter img_bin::Union{Array{Bool,2}, BitArray{2}, Nothing}=nothing to the function? So the interface will be changed from

watershed(img::AbstractArray{T, N}, markers::AbstractArray{S,N})

to

watershed(img::AbstractArray{T, N}, markers::AbstractArray{S,N}, img_bin::Union{AbstractArray{Bool,2}, BitArray{2}, Nothing}=nothing)

On my example, new function works 3.5 times faster. If you think it's fine, I will create a pull request.

[timholy]

This sounds reasonable to me! But perhaps it would be better to make it a keyword argument? Perhaps called mask or something?

[VPetukhov]

But perhaps it would be better to make it a keyword argument? Perhaps called mask or something?

Yep, perhaps it would be better. Thanks!