The application of digital image processing using FFT to monitor web structure

The goal of this project was to develop a new PC-based tool to monitor important structural features in thin nonwoven webs and classify unknown webs based on these features.

First, a 512 X 512 pixel image is snapped by a video camera and displayed on a monitor screen. Then, a 2-D Fourier transform is applied to a fixed area of 256 x 256 pixels to convert data from the spatial domain into the frequency domain; that is, a web image is replaced by a FT spectrum. Energy is accumulated in the FT spectrum in response to the spatial arrangement of fibers over the background. Consequently, analysis of spectral energy provides a feasible path to monitor web structure. Total spectral energy was measured. In addition, the spectral energy distribution was measured using bi-linear interpolation. Several models using simple arrangements of lines were used to study the influence of web structural features on the 2-D FT spectrum. Several indices were extracted from the Fourier spectrum in order to classify webs with regard to structure. This process was illustrated for the case of webs having different fiber orientation and weight where unknown patterns were classified using 3 different training models.