compute_structure_tensor.m

image denoising toolbox in matlab

function H = compute_structure_tensor(M,sigma1,sigma2, options)

% compute_structure_tensor - compute the structure tensor
%
%   T = compute_structure_tensor(M,sigma1,sigma2);
%
%   sigma1 is pre smoothing width (in pixels).
%   sigma2 is post smoothing width (in pixels).
%
%   Follows the ideas of
%   U. K鰐he, "Edge and Junction Detection with an Improved Structure Tensor", 
%   Proc. of 25th DAGM Symposium, Magdeburg 2003, 
%   Lecture Notes in Computer Science 2781, pp. 25-32, Berlin: Springer, 2003
%
%   You can set:
%       - options.sub=2 [default=2] to perform x2 interpolation of the
%           gradient to avoid aliasing.
%       - options.use_renormalization=1 [detault=0] to compute the tensor
%           using unit norm gradient vectors.
%       - options.use_anisotropic=1 [detaul=0] to perform anisotropic 
%           smoothing using a hour-glass shaped gaussian kernel.
%           This better capture edges anisotropy.
%
%   Copyright (c) 2006 Gabriel Peyr