REGIONADJACENCY - Region adjacency matrix of a labelled image.

Contents

Description

Compute a (sparse) adjacency matrix for a labelled image of segmentation.

Syntax

       G = REGIONADJACENCY(L);
       G = REGIONADJACENCY(L, labels, conn);

Inputs

L : a segmentation image, where each pixel is assigned the (integer) label of the region it belongs to.

labels : (optional) labels of interest: the graph will be output for those labels only; default: labels=[], ie. all (non null) labels present in the input image will be represented.

conn : (optional) 4 or 8 connectivity; default: conn=8.

Outputs

G : a (sparse) adjacency matrix indicating (through non zero entries) which regions are adjacent to each other, ie. pixels belonging to those regions share boundaries.

Remarks

See also

Ressembles: LABELMATRIX, BWLABEL, BWCONNCOMP. Requires: REGIONADJACENCY_BASE.

Function implementation

function G = regionadjacency(L, varargin)

parsing parameters

error(nargchk(1, 3, nargin, 'struct'));
error(nargoutchk(1, 1, nargout, 'struct'));

% mandatory parameter
if ~isnumeric(L)
    error('regionadjacency:inputerror','image of integer labels required in entry');
end

p = createParser('REGIONADJACENCY');   % create an instance of the inputParser class.
p.addOptional('labels', [], @(x)isempty(x) || (isnumeric(x) && all(x(:)>=0)));
p.addOptional('conn', 8, @(x)isscalar(x) && (x==4 || x==8));

% parse and validate all input arguments
p.parse(varargin{:});
p = getvarParser(p);

setting variables

if isempty(p.labels)
    % identify the unique labels in the image, excluding 0 as a label
    p.labels = setdiff(unique(L(:)),0);
elseif max(p.labels(:))>max(L(:))
    warning('regionadjacency:inputwarning', ...
        'labels not represented in the input segmentation are ignored');
    p.labels(p.labels(:)>max(L(:))) = [];
end

main computation

G = regionadjacency_base(L, p.labels, p.conn);

end % end of regionadjacency