Multi-scale image analysis for stochastic detection of self-similarity in complex texture

Abstract

A method is presented for detecting self-similarity via multi-scale image analysis. By integrating multi-scale images, the missing probability is represented for unknown attractors. This implies that the collection of local minimum points of the missing probability specifies the stochastic feature of observed patterns. The self-similarity is detected via the design of the imaging process generating the most complex pattern. The method is verified by simulation studies.