Evaluation of fractional Brownian motion synthesis methods using the SVM classifier

With regard to the computer graphics and virtual reality revolutions, there is an increasing need for automated generation of natural textured images. Hence an increased attention is given to fractals and especially to the fractional Brownian motion (fBm) model that can characterize natural phenomena such as landscapes, mountains and clouds. Besides that, the fBm has been used as an efficient model for medical applications, such as breast cancer and bone disease. The aim of this paper is to evaluate different synthesis methods of fBm images. A classification approach is used to evaluate the quality of synthesis methods in terms of adequation of the generated realizations to fBm properties. The Dual-tree MBand Decomposition Transform (DMBDT) is used to extract different statistical features from the generated images to be used in a classification task. Four well known synthesis methods were evaluated. Obtained classification rates show the superiority of the Stein method which is conform to the literature. The classification scheme is then applied to bone X-ray images to distinguish between two groups of osteoporotic (OP) and control cases (CC). An accuracy of 96% was reached proving that the proposed classification approach is also efficient for real clinical applications. (C) 2018 Elsevier Ltd. All rights reserved.