Abstract
A texture descriptor is a collection of quantified measurements of a texture’s properties. They are often used for classifying texture images and recognition of objects in an image that have repeated patterns. In the past, quantified measures have often been calculated in the spatial domain based on the texture pattern, and include Coarseness, Linelikeness and Directionality. These are the local texture features which describe the basic unit of texture that is repeated, whereas global texture features describe how the unit of texture is repeated throughout the image.
This research proposes a new approach for finding descriptors of local and global texture features by calculating them in the frequency domain. The research takes advantage of properties of the Fourier spectrum to introduce a Directional Unification descriptor, which measures how much the lines in a pattern are in the same direction and Texture Pattern Formation feature descriptors, which describe how a pattern repeats spatially within a texture.
In order to discover the advantages and the limitations of the texture feature descriptors in frequency domain, this research uses an experimental methodology. Standard testing images are used to compare between the spatially-based approaches and the frequency-based approaches. Additional images were included to facilitate the investigation of specific texture features. This testing demonstrated and that the new approach provides translation invariance to the descriptors, they are less affected by image intensity and image clarity.
