Unsupervised Image Anomaly Detection and Localization in Industry Based on Self-Updated Memory and Center Clustering

Abstract

Defect detection of industrial products often uses computer vision methods. Detecting anomalies in the image can reflect the defect of the product. In order to adapt to the scene of less defect samples and unclear defect classification standards in industrial production and improve the accuracy and robustness of detection, this paper proposes a new unsupervised anomaly detection and Localization framework based on self-updated memory and center clustering (SMCC). Distinct from previous works, it uses a pre-trained model to extract image features, and then uses a Gaussian mixture model to cluster and obtain cluster centers, so that normal sample features are compactly distributed around the cluster centers, thereby better distinguishing normal and abnormal sample features. The advantage of the self-updated memory bank is to reduce the use of memory and adjust the parameters of the pre-trained network to make it more suitable for the distribution of the current dataset. Our experiments on the MVTec AD and other datasets show the effectiveness of SMCC for anomaly detection and localization.