A Contrario Multi-scale Anomaly Detection Method for Industrial Quality Inspection

AbstractAnomalies can be defined as any non-random structure that deviates from normality. Anomaly detection methods reported in the literature are numerous and diverse, what is considered anomalous usually varies depending on particular scenarios applications. In this work, we propose an a contrario framework to detect anomalies images applying statistical analysis feature maps obtained via convolutions. We evaluate filters learned image under patch PCA, Gabor pre-trained deep neural network (Resnet). The proposed method multi-scale fully unsupervised able wide variety of scenarios. While end goal work subtle defects leather samples for automotive industry, show same algorithm achieves state-of-the-art results public datasets.