We have developed a machine vision method that will be able to detect errors in materials that show a periodic structure. The method is based on a two-dimensional Fourier transform.
Our basic assumption is that the Fourier transformation of the image that exhibits a periodic structure results in a pattern which contains a large number of peaks. Conversely, it contains a small number of peaks when the sample contains a defect, which disrupts the periodic structure. The peaks are detected by the MSER detector. The output of the MSER detector is the number of peaks. To illustrate and evaluate the proposed method, we used textile samples in which we created defects (tearing, puncturing, cutting). We systematically collected a database of images, and marked the defects in textile patterns as polygons. For each polygon we calculated the bounding box that was used in sample extraction from images. Samples were classified to those without defect and those containing a defect. For the detection of defects, we have qualitatively illustrated the operation of the error detector using the presented classifier and the sliding window method.
|
