TENSCALEDIRFILT - Scale adaptive directional Gaussian filtering.

Contents

Description

Perform Gaussian adaptive directional filtering along some tensor field using a local adaptive scale as in [PVS06].

Syntax

   F = TENSCALEDIRFILT(I, T);
   F = TENSCALEDIRFILT(I, T, S);
   F = TENSCALEDIRFILT(I, T, 'Property', propertyvalue, ...);

Inputs

T : either a vector field (matrix of size (n,n,2)) or a tensor field (matrix of size (n,n,2,2)).

Property [propertyname propertyvalues]

aecc : tuning parameter >0 setting an upper-bound on the eccentricity of the filtering; default: aecc=0.5 for a maximum eccentricity of 3 when the anisotropy of the tensor field is 1.

sig1, sig2 : when T is a vector (and not tensor) field, these parameters parse the (locally constant) directional scales of the anisotropic Gaussian kernel: sig1 is the (fixed) scale along the elongated orientation and is greater than or equal to sig2; default: sig1=8 and sig2=3 when T is a vector field; otherwise, note that when T is a tensor, the same variables are estimated (locally) from the eigen- values of T.

Output

F : image filtered using the tensor field given in input.

Reference

[PVS06] T.Q. Pham, L.J. van Vliet, and K. Schutte: "Robust fusion of irregularly sampled data using adaptive normalized convolution", EURASIP Journal on Applied Signal Processing, 2006(1:12), ID 83268, 2006. http://www.hindawi.com/journals/asp/2006/083268/abs/

See also

Ressembles: CONVOLUTION, TENSANIFILT, ADAPTIVEFILT, GEODESICFILT, MDLFILT. Requires: TENSCALEDIRFILT_BASE.

Function implementation

function F = tenscaledirfilt(I,T,varargin)

parsing and checking parameters

error(nargchk(2, 13, nargin, 'struct'));
error(nargoutchk(1, 1, nargout, 'struct'));

if ~isnumeric(I)
    error('tenscaledirfilt:errorinput','a matrix is required in input');
elseif size(T,4)>2
    error('tenscaledirfilt:errorinput','T must be a tensor field or a vector field');
end

p = createParser('TENSCALEDIRFILT');
p.addOptional('S',[], @(x)isnumeric(x) && all(x(:)>0));
p.addParamValue('sig1',8, @(x)isscalar(x) && x>0);
p.addParamValue('sig2',3, @(x)isscalar(x) && x>0);
p.addParamValue('nthe', 12, @(x)isscalar(x) && round(x)==x);
p.addParamValue('nsig', 5, @(x)isscalar(x) && round(x)==x);
p.addParamValue('nani', 4, @(x)isscalar(x) && round(x)==x);
p.addParamValue('aecc', 0.5, @(x)isscalar(x) && round(x)==x);
p.addParamValue('ani', false, @(x)islogical(x));

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

main computation

F = tenscaledirfilt_base(I, T, p.S, p.sig1, p.sig2, ...
    p.nsig, p.nthe, p.nani, p.aecc, p.ani);

if p.disp
    figure, imagesc(rescale(F,0,1)), axis image off;
    if size(F,3)==1, colormap gray; end;
    title('adaptively filtered image');
end

end % end of tenscaledirfilt