Texture structure analysis via description of primitive configuration and repeating pattern

摘要

This paper addresses the problem of describing the primitive structure and the repeating pattern in texture images. Effective methods exist for statistical analysis for texture recognition and segmentation. However, the statistical descriptions can hardly reveal the information on the structure of primitives and how they repeat in the texture image. A method for extracting structure features of texture image is proposed in this paper: based on an analysis of edge features in scale space, the location and orientation of edges have been extracted. At coarse scale, edges from the contour of the same primitive are determined to be linked together by a criteria defined as linking strength, thus the localization of each primitive is determined. With the corresponding edges at fine scale, the parameterization of primitive configuration is optimized with parameter space determined repeating patterns as constraints.