Abstract

The number and shapes of primitives in multiprimitive textures are estimated using a texture generation model that various grains derived from primitives are randomly arranged. This method segments a texture image into each grains by the watershed method and locate each grain in a feature space based on the morphological size distribution. Although shapes of the segmented grains vary, the grains derived from a common primitive are located in a neighborhood in the feature space. The number and shapes of primitives are derived by extracting clusters of grains in the feature space. The principal component analysis is applied to discriminating clusters clearly.

Keywords

texture analysis, mathematical morphology, multivariate analysis, principal component analysis, cluster analysis