LOCALGLCM2D - Textural features based on the Grey Level Co-occurrence Matrix.

Description
Syntax
Inputs
Output
References
See also
Function implementation

Description

Compute local textural features using the 2D histograms of joint distribution of greylevel pairs as defined in the Grey Level Co-occurrence Matrix (GLCM) approach of [Hara79,HSD73].

Syntax

   O = LOCALGLCM2D(I);
   O = LOCALGLCM2D(I, 'Property', propertyvalue, ...);

Inputs

I : input image with dimension C, possibly multispectral (C>1 bands).

Properties [propertyname propertyvalues]: 'dcar' : (nd,2) matrix list of nd displacement vectors, expressed in cartesian coordinates X-Y, used for the estimation of the 2D joint greylevel distributions; default: nd=1 and dcar=[1 1].

'dpol' : similarly, (nd,2) matrix list of nd displacement vector expressed in polar coordinates; incompatible with previous 'dcar'; default: ignored, 'dcar' is used.

'win' : fixed size of the analyzing window used for local estimation;

'res' : number of levels of quantization of the image (ie. the resolution of the histograms); if is set to 0, the resolution will be a constant estimated w.r.t. to the window size; default: res=64.

'wei' : parameter defining the weighting scheme adopted for the estimation of local neighbourhoods used in histogram weighting; it is either:

'ave' for uniform averaging weighting function,
'gaus' for Gaussian weighting function: the neighbourhood for estimating the histogram distribution is weightened by a Gaussian function following the approach of [JC06],
'inv' for multiscale weighting function: the neighbourhood is weightened by the inverse euclidean distance to the central pixel following the approach of [GTY02,GTY03];

default: wei='inv'.

'sig' : optional parameter defining either:

the standard deviation of the Gaussian window when wei='gaus',
the scaling factor of the euclidean window when wei='inv', depending on the option 'wei' selected above;

default (in both cases): sig=2.

'mask': optional label image of categorisation (with values >=0).

'im' : optional label(s) for which the textural features are computed when the image of labels mask has been passed as well; default=-1, ie. the textural features are estimated for all pixels in the input image.

Output

O : GLCM-based features.

References

[HSD73] R.M. Haralick, K. Shanmugam, and I. Dinstein: "Textural features for image classification", IEEE Trans. Systems, Man and Cybernetics, 3(6):610-621, 1973. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4309314

[Hara79] R.M. Haralick: "Statistical and structural approaches to texture", Proceedings of IEEE, 67:786-804, 1979. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1455597

[ST99] L. Soh and C. Tsatsoulis: "Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices", IEEE Trans. on Geoscience and Remote Sensing, 37(2), 1999. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=752194

[Clausi02] D A. Clausi: "An analysis of co-occurrence texture statistics as a function of grey level quantization", Canadian Journal of Remote Sensing, 28(1):45-62, 2002. http://www.eng.uwaterloo.ca/~dclausi/Papers/Published%202002/Clausi%20-%20GLCP%20and%20quantization%20-%20CJRS%202002.pdf

[GTY02] J. Grazzini, A. Turiel, and H. Yahia: "Entropy estimation and multiscale processing in meteorological satellite images", Proc. of ICPR, vol. 3, pp: 764-768, 2002. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1048103

[GTY03] J. Grazzini, A. Turiel, and H. Yahia: "Analysis and comparison of functional dependencies of multiscale textural features on monospectral infrared images", Proc. of IGARSS, vol. 3, pp: 2045-2047, 2003. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1294334

[JC06] R. Jobanputra and D.A. Clausi: "Preserving boundaries for image texture segmentation using grey level co-occurring probabilities", Pattern Recognition, 39:234-245, 2006. http://www.sciencedirect.com/science/article/pii/S0031320305002992

Function implementation

function O = localglcm2d(I, varargin)

parsing parameters

if ~isnumeric(I)
    disp('a matrix is required in input'); return;
end

% optional parameters
p = createParser('LOCALGLCM2D');   % create an instance of the inputParser class.
% mandatory (required) variables
%p.addRequired('I', @isnumeric);
p.addParamValue('res', 64, @(x)isscalar(x) && x>=0);
p.addParamValue('dcar', [1 1], @(x)isnumeric(x) && min(size(x))<=2 && ....
    all(round(x(:))==x(:)));
p.addParamValue('dpol', [], @(x)isnumeric(x) && any(size(x))==2 && ...
    all(x(:,1))>0 && -pi<all(x(:,2)) && pi>all(x(2)));
p.addParamValue('win', [], @(x)isscalar(x) && x>=1);
p.addParamValue('wei', 'ave', @(x)ischar(x) && ...
    any(strcmpi(x,{'gaus','ave','inv'})));
%p.addParamValue('n', 'global', @(x)ischar(x) && ...
%    any(strcmpi(x,{'local','global'})));
p.addParamValue('sig', '2.0', @(x)isscalar(x) && x>0.1 && x<10);
p.addParamValue('mask', [], @(x)isnumeric(x) && all(x>=0));
p.addParamValue('feat', 'all', @(x)ischar(x) && ...
    any(strcmpi(x,{'all','con','ene','ent','ent10',...
    'hom','var','dis','cor','inv'})));

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

checking parameters and setting internal variables

if strcmp(p.wei,'gaus')
    if ~isempty(p.sig) && isempty(p.win)
        p.win = 6*p.sig + 1;
    end
elseif isempty(p.win)
    p.win = 7;
end

% ensure that the window size is odd
if ~isempty(p.win) && mod(p.win,2) == 0
    p.win = p.win + 1;
end

main calculation

O = localglcm2d_base(I, p.feat, p.res, p.dcar, p.dpol, p.win, p.wei, p.sig, p.mask);

end % end of localglcm2d