The construction of texton-based texture

classifiers consists of three main stages: (1) construction of a texton codebook, (2) computation of a texton

frequency histogram, and (3) training of the classifier based on the texture frequency histograms.

texton codebook contain  small texture patches are extracted from random positions in the texture images. size of texton will depend upon types of textures available in image and size of original image texton size should be small enough to make cluster of all available textures in image

pls refere to the files

Hi,

the question is what you consider as a texton. It can be either pixel with some surrounding in variable domains, or a patch in texture containing smallelst repeatable element in material structure.

The former was covered in previous reply, for the latter: For regular materials I would look at frequency peaks in original image FFT domain to get approximate size of the texton. Then you can use image tiiling approach - i.e. cutting of representative tile from original texture image containg the texton. If the tile (texton) size and location is estimated correctly, it allow synthesis of original texture image just by replicating the texton according to a precomputed index.

See e.g. here:

Haindl Michal, Hatka Martin : Generalized Roller , Proceedings of the 17th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, WSCG 2009, p. 73-80 ,

Somol Petr, Haindl Michal : Novel path search algorithm for image stitching and advanced texture tiling , WSCG'2005. 13th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision 2005. Proceedings, p. 155-162