merge region based on area

Question

i need to merge small regions from the Label product of scikit-image slic until reach an area regions threshold

the slic itself has a min_size_factor which is not the solution in my case because it has unwanted affect at overall result of labeling.

the way i can calc regions area :

def pixelcount(regionmask):
    return np.sum(regionmask)
props = regionprops(label_img, extra_properties=(pixelcount,))

there is a way same as cut_normalized function (which work on colors similarity of pre-calculated RAG) for regions area?

any help appreciated.

Answer 1

for someone who may encounter this :

• build a RAG by scikit_image with "pixel count" as data attached to nodes

• define "weight" equal to below code and merge normalize :

    for x, y, d in graph.edges(data=True):
        diff = graph.nodes[x]['pixel count'] - graph.nodes[y]['pixel count']
        diff = np.linalg.norm(diff)