Function to fill interior holes

[mkitti]

I would like to add a function that fills interior holes in binary images.

This is roughly equivalent to MATLAB's imfill with the 'holes' option specified.

This may also be similar to https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.binary_fill_holes.html

This is distinct from the current imfill here which uses a flood fill algorithm.

The algorithm to fill interior, black, holes is as follows.

1. Find black pixels on the boundary of the image
2. Invert the image so that holes and background are black, false.
3. Use label_components to label the holes and background regions.
4. Identify components that contain the black pixels on the boundary image identified in Step 1.
5. The resulting image fills all the components containing black pixels on the boundary with black. All other pixels are white.

Here's a quick implementation.

julia> function imfill_interior_holes(I)
           boundary_pixels = filter(ci->(any(ci.I .∈ indices[1].I) || any(ci.I .∈ indices[end].I)) && !I[ci], indices)
           labels = label_components(.!I)
           background_labels = unique(labels[boundary_pixels])
           return labels .∉ Ref(background_labels)
       end
imfill_interior_holes (generic function with 1 method)

julia> I = falses(8,8); I[1:8,3:6] .= 1; I[[CartesianIndex(4,4), CartesianIndex(5,5)]] .= 0; I
8×8 BitMatrix:
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  0  1  1  0  0
 0  0  1  1  0  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0

julia> imfill_interior_holes(I)
8×8 BitMatrix:
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0

This a breakdown of the steps.

julia> I = falses(8,8); I[1:8,3:6] .= 1; I[[CartesianIndex(4,4), CartesianIndex(5,5)]] .= 0; I
8×8 BitMatrix:
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  0  1  1  0  0
 0  0  1  1  0  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0

julia> boundary_pixels = filter(ci->(any(ci.I .∈ indices[1].I) || any(ci.I .∈ indices[end].I)) && !I[ci], indices);

julia> labels = label_components(.!I)
8×8 Matrix{Int64}:
 1  1  0  0  0  0  4  4
 1  1  0  0  0  0  4  4
 1  1  0  0  0  0  4  4
 1  1  0  2  0  0  4  4
 1  1  0  0  3  0  4  4
 1  1  0  0  0  0  4  4
 1  1  0  0  0  0  4  4
 1  1  0  0  0  0  4  4

julia> background_labels = unique(labels[boundary_pixels])
2-element Vector{Int64}:
 1
 4

julia> labels .∉ Ref(background_labels)
8×8 BitMatrix:
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0
 0  0  1  1  1  1  0  0

[ThomasRetornaz]

Hello o/ the generic and simplest way to achieve that is to use morphologycal reconstruction (mreconstruct in julia ops)
see img https://www.researchgate.net/figure/Fill-holes-transformation-Morphological-reconstruction-by-erosion-of-f-from-f-marker_fig2_259134752

[mkitti]

Hello o/ the generic and simplest way to achieve that is to use morphologycal reconstruction (mreconstruct in julia ops) see img https://www.researchgate.net/figure/Fill-holes-transformation-Morphological-reconstruction-by-erosion-of-f-from-f-marker_fig2_259134752

That is one approach to the gray scale case.

[ThomasRetornaz]

The implem is generic see #119
May we could add a specific implem for binary case and/or specialize mreconstruct for binary case