Anomaly detection method near arbitrary boundary field for surface inspection

Abstract

Surface inspection of products has become an essential factor in improving appearance quality. Surface inspection is performed using the brightness deviation from the reference images or the neighboring pixels. If there is no anomaly, there is no deviation, but if there are anomalies, the deviation is generated. Detecting the anomaly on an object's original shape is not easy. In this paper, to solve this problem, we propose a method of extracting the information of perpendicular norm at the boundary which is the shape information of the target object, and generating a transformed two-dimensional(2D) image using the pixel interpolation method. The transformed 2D image information is used as data that efficiently extracts anomalies. Furthermore, if an area in which a boundary can be extracted, anomalies can be detected by making a transformed 2D image of the neighboring region of the boundary. In order to verify the performance of the proposed method, experiments were conducted based on objects with anomalies on various shapes. It is necessary to acquire a high-resolution image of UHD or higher to detect detailed anomalies. We expect the anomaly detection algorithm proposed in this paper to enable surface inspection of areas that were difficult to inspect.