An adaptive threshold segmentation method based on BP neural network for paper defect detection

Abstract

Threshold segmentation is the fastest method of defect detection in the modern defect inspection system based on computer vision. But in the real paper defect detection system, the segmentation thresholds usually change with the paper image luminance which is influenced by many factors. In order to resolve this problem, an adaptive threshold segmentation method based on BP neural network is proposed in this paper. For this method, BP neural network models are created and trained to obtain the segmentation thresholds according to the image luminance and the defects are segmented with these thresholds obtained by the network. This method is especially suitable for detecting three typical types of paper defects: dark spot, light spot and hole. The experiment results indicate that this method is efficient and can be applied to modern paper defect inspection system.