Image texture features and multivariate image analysis (MIA)

Texture may be described as a pattern that is spatially repeated either deterministically or stochastically.

Image texture analysis

In images this pattern consists of pixels representing the “true nature”. Textures are classified on a continuum from isotropic (having no particular orientation) to strongly anisotropic. Granite, for example, is isotropic, whereas a texture comprised of horizontal layers of sandstone is anisotropic.  In order to compute the classical texture measures there is often a need to perform extensive calculations on the images and to pre-process them in a specific manner. Some of these texture measures are constructed to estimate specific information. Other texture measures seem to be more global in nature. These approaches are closely linked to a physical understanding of the texture measures and are often handled by univariate techniques. By combining several feature extraction techniques our approaches are typically multivariate in the sense that they extract a vector of several features from each image. Multivariate statistical analysis is therefore required in order to understand the relationship among the variables and to find relationships to external physical and other related features. We develop methods to help in this process. A reference to our work on AMT (Angle Measure Technique) is found here. http://arken.umb.no/~kkvaal/eamtexplorer
Contact: Knut Kvaal (knut.kvaal@nmbu.no)

Bi-plot from a PCA analysis of Brodatz textures
Bi-plot from a PCA analysis of Brodatz textures Photo: Knut Kvaal

 

Published 8. April 2014 - 15:17 - Updated 23. May 2017 - 19:42