AMOEBASUPERPIX - Amoeba-like superpixel segmentation.

Contents

Description

Compute superpixel regions using an algorithm based on the estimation of local amoeba-like neighborhoods. This approach is similar to the Simple Linear Iterative Clustering superpixel segmentation technique of [ASSLFS10,LSALF10] and introducing a distance derived from the amoeba construction in [LDM0709]. It performs a kmeans-based clustering for a selected number of seeds that outputs superpixel regions as amoeba neighbourhoods [LDM0709,GS08].

Syntax

  Q = AMOEBASUPERPIX(I);
  [Q, Ck, ColCk] = AMOEBASUPERPIX(I, K0, 'Property', propertyvalue, ... );

Inputs

I : input color image of size (X,Y,C) (multispectral with C=3 bands).

K0 : desired number of clusters, ie. of approximately equally-sized superpixels.

Property [propertyname propertyvalues]

'thres' : (optional) stopping criterion; it is defined as a threshold on the error for relocating all the superpixel regions' centers; default: thres=eps.

'alpha' : (optional) vector of dimension (1,n) with n=1 or 2 providing the influence factors used when propagating the amoeba superpixels; it quantifies the relative influences of the spectral proximity w.r.t the spatial one (simply computed as the local Euclidean distance) in the calculation of the perceptual distance measure; alpha(1) is a scalar used to amplify (multiply) the spectral (graylevel or color) influence (the higher it is, the more spectral closeness is emphasized), alpha(2) is a scalar used to introduce a factor based on the gradient magnitude (the higher it is, the less probable amoeba superpixels will cross regions of high gradient); note that the higher alpha(1) and/or alpha(2), the less compact the clusters are; default: alpha=[1 0].

'n' : (optional) multiplying factor used for the search area around each superpixel region's center; default: n=2.

'k' : (optional) size of the neighbourood considered when correcting the location of the superpixel regions' centers; when set to 0, the first initial initialization is kept as it is; default: k=3.

'iter' : (optional) maximum number of iterations; default: iter=Inf, ie. the segmentation process is iterated till convergence.

'lab' : (optional) boolean flag set to true for transforming, prior to the processing, a 3D image (assumed to be RGB) into the Lab color space; default: lab=false.

Outputs

Q : diagram of amoeba superpixels regions; it takes values in the range [1,K] where K is the final number of superpixel regions; it is a matrix of size (X,Y).

Ck : coordinates of the centers of the corresponding superpixels; it is a matrix of size (K,2).

ColCk : representative color values of the superpixels; it is a matrix of size (K,3).

References

[LDM0709] R. Lerallut, E. Decenciere, and F. Meyer: "Image filtering using morphological amoebas", Image and Vision Computing, 25(4):395-404, 2007. http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.106.1413

[GS08] J. Grazzini and P. Soille: "Adaptive morphological filters using similarities based on geodesic time", Proc. DGCI, LNCS, vol. 4992, pp. 519-528, 2008. http://www.springerlink.com/content/f6v62233xqkklq72/

[GSD10] J. Grazzini, P. Soille and S. Dillard: "Multichannel image regularisation using anisotropic geodesic filtering", Proc. ICPR, pp. 2664-2667, 2010. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5596008

[ASSLFS10] R. Achanta, A. Shaji, K. Smith, A. Lucchi, P. Fua and S. Susstrunk: "SLIC superpixels", EPFL Technical Report no. 149300, 2010. http://infoscience.epfl.ch/record/149300/files/SLIC_Superpixels_TR_2.pdf

[LSALF10] A. Lucchi, K. Smith, R. Achanta, V. Lepetit and P. Fua: "A fully automated approach to segmentation of irregularly shaped cellular structures in EM images", pp. 463-471, Proc. MICCAI, 2010. http://cvlab.epfl.ch/publications/publications/2010/LucchiSALF10.pdf

See also

Ressembles: SLICSUPERPIX, GEOSUPERPIX, FMM. Requires: AMOEBASUPERPIX_BASE.

Function implementation

function [Q, varargout] = amoebasuperpix(I, varargin )

parsing parameters

narginchk(1, 23);
nargoutchk(1, 3);

% mandatory parameter
if ~isnumeric(I)
    error('amoebasuperpix:inputerror','a matrix is required in input');
end

% optional parameters
p = createParser('AMOEBASUPERPIX');   % create an instance of the inputParser class.
p.addOptional('K', 0.1, @(x)isscalar(x) && ((x>0 && x<1) || (x>1 && x==round(x))));
% additional optional parameters
p.addParamValue('alpha', [1 0], @(x) all(x>=0) && (isscalar(x) || ...
    (isnumeric(x) && length(x)<=2)));
p.addParamValue('thres', eps, @(x) isscalar(x) && x>=0);
p.addParamValue('n', 2, @(x)isscalar(x) && x>=1 && x==round(x));
p.addParamValue('k', 0, @(x)isscalar(x) && (x==0 || x>=3));
p.addParamValue('iter', Inf, @(x)isscalar(x) && x>=1);
p.addParamValue('lab', false, @islogical);
% additional parameters used by IM2POTENTIAL_BASE

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

setting variables

if size(I,3)~=3 && p.lab
    warning('amoebasuperpix:inputwarning', ...
        'option lab available for RGB and 3 vectorial image only');
    p.lab = false;
end

if p.K<1
    p.K = p.K * min(size(I,1),size(I,2)); % proportion
end

% fill p.alpha with 0's
if length(p.alpha)<2,  p.alpha = [p.alpha 0];  end

main computation

[Q, Ck, ColCk] = amoebasuperpix_base(I, p.K, ...
    p.lab, p.alpha, p.thres, p.n, p.k, p.iter);

if any(~Q(:))
    i = find(~Q);
    warning('amoebasuperpix:outputwarning', ...
        ['not all pixels reached by the classifier - ' ...
        'increase the range of the exploration domain (''n'') ' ...
        'or, when possible, the number of iterations (''iter'')']);
    Q(i) = 1;
end

if nargout>=2
    varargout{1} = Ck;
    if nargout>=3,  varargout{2} = ColCk;  end
end

display

if p.disp
    figure;
    if isempty(ver('images'))
        subplot(1,2,1), imagesc(rescale(I));
        subplot(1,2,2), imagesc(Q), colormap jet;
    else
        M = (imdilate(Q,ones(3,3))-Q==0);
        subplot(1,2,2), imagesc(label2rgb(Q.*M)), axis image off;
        M = cat(3,M,M,M);
        subplot(1,2,1), imagesc(rescale(I.*M)+(1-M)), axis image off;
    end
    suptitle('amoeba superpixel regions');
    figure;
    subplot(1,2,1), imagesc(rescale(I)), axis image off;
    if p.lab,  ColCk = Lab2RGB(ColCk(:,1),ColCk(:,2),ColCk(:,3));  end
    ColCk = double(reshape(ColCk(Q(:),:), size(I)));
    if exist('i','var'),  ColCk([i; i+numel(Q); i+2*numel(Q)]) = 0;  end
    subplot(1,2,2), imagesc(rescale(ColCk)), axis image off;
    suptitle('amoeba superpixel approximation');
end

end % end of amoebasuperpix