Classification of periodic patterns using hough transform

Abstract

Texture is an important characteristic in an image and it is used to identify objects or regions of interest within an image. The problem of texture classification has been widely studied and most of the proposed techniques fall into two categories; statistical or structural. In this paper a structural approach based on the Hough method for line detection is proposed, for the description of periodic textures that consist of mostly straight lines. The features used for classification procedure are based on the relative orientations and separations of lines. With the proper normalizations, classification is independent of geometrical transformations such as rotation, translation, and/or scaling. This approach provides more unique features compared to other approaches. Experimental results with test patterns are presented.