% CLEANUPREGIONS Cleans up small segments in an image of segmented regions
%
% Usage: [seg, Am] = cleanupregions(seg, areaThresh, connectivity)
%
% Arguments: seg - A region segmented image, such as might be produced by a
% graph cut algorithm. All pixels in each region are labeled
% by an integer.
% areaThresh - Regions below this area in pixels will be merged with an
% adjacent segment. I find a value of about 1/20th of the
% expected mean segment area, or 1/1000th of the image area
% usually looks 'about right'.
% connectivity - Specify 8 or 4 connectivity. If not specified 8
% connectivity is assumed.
%
% Note that regions with a label of 0 are ignored. These are treated as
% privileged 'background regions' and are untouched. If you want these
% regions to be considered you should assign a new positive label to these
% areas using, say
% >> L(L==0) = max(L(:)) + 1;
%
% Returns: seg - The updated segment image.
% Am - Adjacency matrix of segments. Am(i, j) indicates whether
% segments labeled i and j are connected/adjacent
%
% Typical application:
% If a graph cut or superpixel algorithm fails to converge stray segments
% can be left in the result. This function tries to clean things up by:
% 1) Checking there is only one region for each segment label. If there is more
% than one region they are given unique labels.
% 2) Eliminating regions below a specified size and assigning them a label of an
% adjacent region.
%
% See also: MCLEANUPREGIONS, REGIONADJACENCY, RENUMBERREGIONS
% Copyright (c) 2010 Peter Kovesi
% www.peterkovesi.com
%
% Permission is hereby granted, free of charge, to any person obtaining a copy
% of this software and associated documentation files (the "Software"), to deal
% in the Software without restriction, subject to the following conditions:
%
% The above copyright notice and this permission notice shall be included in
% all copies or substantial portions of the Software.
%
% The Software is provided "as is", without warranty of any kind.
%
% October 2010 - Original version
% February 2013 - Connectivity choice, function returns adjacency matrix.
function [seg, Am] = cleanupregions(seg, areaThresh, connectivity, prioritySeg)
if ~exist('connectivity','var'), connectivity = 8; end
if ~exist('prioritySeg','var'), prioritySeg = -1; end
% 1) Ensure every segment is distinct but do not touch segments with a
% label of 0
labels = unique(seg(:))';
maxlabel = max(labels);
labels = setdiff(labels,0); % Remove 0 from the label list
for l = labels
[bl,num] = bwlabel(seg==l, connectivity);
if num > 1 % We have more than one region with the same label
for n = 2:num
maxlabel = maxlabel+1; % Generate a new label
seg(bl==n) = maxlabel; % and assign to this segment
end
end
end
if areaThresh
% 2) Merge segments with small areas
stat = regionprops(seg,'area'); % Get segment areas
area = cat(1, stat.Area);
Am = regionadjacency(seg); % Get adjacency matrix
labels = unique(seg(:))';
labels = setdiff(labels,0); % Remove 0 from the label list
for n = labels
if ~isnan(area(n)) && area(n) < areaThresh
% Find regions adjacent to n and keep merging with the first element
% in the adjacency list until we obtain an area >= areaThresh,
% or we run out of regions to merge.
ind = find(Am(n,:));
while ~isempty(ind) && area(n) < areaThresh
if ismember(prioritySeg, ind)
[seg, Am, area] = mergeregions(n, prioritySeg, seg, Am, area);
prioritySeg = n;
else
[seg, Am, area] = mergeregions(n, ind(1), seg, Am, area);
end
ind = find(Am(n,:)); % (The adjacency matrix will have changed)
end
end
end
end
% 3) As some regions will have been absorbed into others and no longer exist
% we now renumber the regions so that they sequentially increase from 1.
% We also need to reconstruct the adjacency matrix to reflect the reduced
% number of regions and their relabeling
[seg, minLabel, maxLabel] = renumberregions(seg);
Am = regionadjacency(seg);
%-------------------------------------------------------------------
% Function to merge segment s2 into s1
% The segment image, Adjacency matrix and area arrays are updated.
% We could make this a nested function for efficiency but then it would not
% run under Octave.
function [seg, Am, area] = mergeregions(s1, s2, seg, Am, area)
if s1==s2
fprintf('s1 == s2!\n')
return
end
% The area of s1 is now that of s1 and s2
area(s1) = area(s1)+area(s2);
area(s2) = NaN;
% s1 inherits the adjacancy matrix entries of s2
Am(s1,:) = Am(s1,:) | Am(s2,:);
Am(:,s1) = Am(:,s1) | Am(:,s2);
Am(s1,s1) = 0; % Ensure s1 is not connected to itself
% Disconnect s2 from the adjacency matrix
Am(s2,:) = 0;
Am(:,s2) = 0;
% Relabel s2 with s1 in the segment image
seg(seg==s2) = s1;