Pattern Fabric Defect Detection Using Nonparametric Regression

Abstract

In this paper we proposed a defect detection on the pattern pabric. This defect usually occured since the pattern is shifted, broken, or different to the intended design. To solve this problem, we use comparing signals approch.To do that, we first modeled the images in a 2D nonparametric regression. Moreover, to cope with the correlation between the neighbour of each pixels, we model the errors correlated in their neighbourhoods. The image is divided into two parts vertically, and then a hypothesis test is constructed so that the left image is the same to the right one, against they are significantly different. Similarly, then the image is divided horizontally, and perform the same test for the upper and lower images. To reject or faild to reject the hypothesis we need to measure the distance between two nonparametric regressions. Since the distribution of test statistic under the hypothesis null is not known. In this method we use the standardized modification of the Mallows distance and construct spatial bootstrap for representing the distribution of the test statistic. To preserve the bound of a pixel to its neighbourhood we construct a spatial bootstrap. The reject hypothesis imply that the defects are found in that particular area.