Search code examples
pythonimage-processingmatrixdisjoint-sets

Find all disjoint subsets of a binray matrix in Pyhton


I have a binary matrix, and I want to find all disjoint subsets that exist in this matrix. To clarify the problem, the matrix is a collection of image masks, masks of irregular shapes, and each disjoint subset of 1s representing a separate mask. In other words, if I have a collection of N disjoint subsets in a matrix of size d1xd2, I want to end up with N matrices of size d1xd2, each having only one of the disjoint subsets of 1s. Any help on that is highly appreciated.


Solution

  • In image processing, what you are trying to do is also called Connected Component Labelling (CCL). Woodford's suggestion of Flood fill algorithm is one algorithm for performing CCL.

    A variation of your question is already answered in this other thread, which uses the function scipy.ndimage.label.
    Here's a slight modification of that answer, which should give you exactly what you are looking for:

    import numpy as np
    from scipy.ndimage import label
    
    # Define your matrix here
    X = np.array([
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
    ], dtype=int)
    
    # How do you define "connected" or "disjointed"? 4-neighbors vs 8-neighbors.
    # https://en.wikipedia.org/wiki/Connected-component_labeling
    structure = np.ones((3, 3), dtype=int)
    
    # Perform connected component labelling
    labeled, ncomponents = label(X, structure)
    
    # Iterate through each labelled blob
    for ii in range(1, ncomponents+1):
        output = np.zeros(X.shape)
        output[labeled == ii] = 1
    
        # Do something to each resulting blob...
        print(output)