Liang K-H, Tjahjadi T & Yang K-H

Proc 1996 IEEE Nordic Signal Processing Symposium, 1996, 239-242

Abstract

Texture Segmentation is a multiscale task because both the global and the local information of an image are required. The Multiscale Texture Segmentation (MTS) employs the Boltzmann probability to determine the change of state P(state) of simulated annealing. This paper presents three modifications of MTS: first, the first two items of the Maclaurin series of the Boltzmann probability is used to determine P(state). Second, the energy increment of P(state) is re-defined to provide a more balanced encouragement of the splitting and merging of a leaf node of the quad-tree used to represent the image. Third, the horizontal and the vertical moments of an image spectrum is used to characterise four natural textures.