Sorry, May be I had send one, But I am not sure if I send suceessfully, because of the attachment is too large, so I send again, with a jpg format.

 Hi every one:

    As I mentioned befor, In the official example, use watershed to segment two circle, In this example, use pick_local_max to find the seed. But When I try on other binary images, there always are two local_max nearby, then the segment be too trivial. And I use the h_maxma and h_minma, I have some trouble:

1. when a region has more than one min value(same but not nearby), the result will mark them all. (I want to use them as seed to watershed, So I need any one, else the region will be broken, can we add a parameter to control?)

2. h_minma run too slowly, (what more, the watershed run more slowly... so my program crashed! I had heard Juan will improve it in the next version)

So I wrote h_minma/maxma and watershed myself (as a sub package of my project imagepy). My watershed now only support 2D, But the h_minma now is OK, works on nd. I had test one image:

       scikit-image's h_minma runs in 15s, and mine runs in 0.3s, 45times.

I will attach my code and test image. If any one can have a look, and improve the h_minma function? I will write a issue on github soon.

                                                                            YXDragon