Abstract
A class of textures are modeled as signals from ‘strewing processes’ in which elements with a certain pattern, or ‘pattern elements’, are strewn or heaped in succession. On the basis of the theoretical consideration of the model, a method is proposed for extracting features of pattern elements constructing the textures. In conventional approaches, random textures have often been taken as Gaussian process and analyzed based on the 2nd-order statistics. On the contrary, because strewing-textures are non-Gaussian, some higher order statistics may contain significant information.
In this paper, a method for analyzing such textures is proposed based on the 3rd-order autocorrelation function in three specific cross-sectional planes, by which one can guess whether the shape of pattern elements is bar-like, disk-like, ring, or periodical. The results of experiments on typical strewing-textures and heaped chip images show the effectiveness of the method.
