Neural fractal prediction of three dimensional surface roughness

This paper presents a methodology for using the high resolution three dimensional (3D) surface data of fabric samples to acquire their surface roughness parameter measurement. Firstly, we compute a parameter FDFFT, which is the fractal dimension estimated from the two-dimensional fast Fourier transform (2DFFT) of 3D surface scan. We validate the rotation-invariance and scale-invariance of FDFFT using fractal Brownian images. Secondly, in order to evaluate the correctness of FDFFT, we provide a method of calculating standard roughness parameters from 3D fabric surface. According to the test results, we demonstrated that FDFFT is a fast and reliable parameter for fabric roughness measurement based on 3D surface data. Finally, we attempt a neural network model using back propagation algorithm and FDFFT for predicting the standard roughness parameters. The proposed neural network model shows good performance to both training samples and test samples.