Generation of Synthetic Multifractal Realistic Surfaces Based on Natural Model and Lognormal Cascade: Application to MRI Classification

摘要

This paper presents a method of generating realistic synthetic multi-fractals surfaces, constructed with multiplicative cascades, that follow lognormal probability density function. The conservation of the natural image gradient direction, and the variance of the difference minimization at each scale between natural image multipliers and those of the selected lognormal model, preserves the initial texture structure. Validation of the model is made with wavelet leader based multifractal analysis, we also propose an application to MRI classification of trabecular bone texture, to differentiate between healthy and osteoporotic cases.