Average Blurring-based Anomaly Detection for Vision-based Mask Inspection Systems

Abstract

When facial masks are produced, various types of defects may appear on mask filters. These defects may include the hair of the inspectors and unexpected raw materials in the production processes. This paper proposes a new method for detecting anomaly regardless of the size and shape of defects. The proposed method uses two-step image processing to detect anomaly. The first step is to use Average Blurring on the mask filter image for image blurring. The most important thing in this step is the kernel size of the Average Blurring is increased to extend the pixel value with defects to the surrounding pixels. In the second step, the Pearson correlation coefficient between the normal mask filter image and the input mask filter image is used according to kernel size. The larger the kernel size of Average Blurring, the lower their correlation coefficient. If the correlation coefficient at a particular kernel size is lower than the threshold value, it is decided as defective image.