Following this thread on the mailing list, I initiate this issue to enhance the peak detection in scikit image.

Currently, we have two functions:

• skimage.morphology.local_maxima
• skimage.feature.peak_local_max

In particular, peak_local_max (and derivatives) is used in scikit-image at different places.

Our functions seem to be sub-optimal and lead to questionable peak detection, especially for plateaus as pointed out on the mailing list. It is also questioned to keep two separate functions.

Recently, @lagru committed to scipy scipy/scipy#8264 a new function to detect peaks based on their prominence. This function is particularly easy to use and efficient for 1D data. Then, it would be worth to consider whether it can be generalized to 2D data. @lagru, would you like to share your expertise on this and eventually contribute to scikit-image if you wish and if it makes sense to you?

Thanks.