Reconstruction by inpainting for visual anomaly detection

•A reconstruction-by-inpainting-based anomaly detection method (RIAD) was proposed.•RIAD achieves state-of-the-art performance on anomaly detection and localization.•We compare RIAD anomaly detection results with recent anomaly detection methods.•The generality of RIAD is demonstrated by applying it...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Pattern recognition Jg. 112; S. 107706
Hauptverfasser: Zavrtanik, Vitjan, Kristan, Matej, Skočaj, Danijel
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.04.2021
Schlagworte:
Anomaly detection
CNN
Inpainting
Video anomaly detection
Video anomaly detection
CNN
Inpainting
Anomaly detection
ISSN:0031-3203, 1873-5142
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•A reconstruction-by-inpainting-based anomaly detection method (RIAD) was proposed.•RIAD achieves state-of-the-art performance on anomaly detection and localization.•We compare RIAD anomaly detection results with recent anomaly detection methods.•The generality of RIAD is demonstrated by applying it on video anomaly detection. Visual anomaly detection addresses the problem of classification or localization of regions in an image that deviate from their normal appearance. A popular approach trains an auto-encoder on anomaly-free images and performs anomaly detection by calculating the difference between the input and the reconstructed image. This approach assumes that the auto-encoder will be unable to accurately reconstruct anomalous regions. But in practice neural networks generalize well even to anomalies and reconstruct them sufficiently well, thus reducing the detection capabilities. Accurate reconstruction is far less likely if the anomaly pixels were not visible to the auto-encoder. We thus cast anomaly detection as a self-supervised reconstruction-by-inpainting problem. Our approach (RIAD) randomly removes partial image regions and reconstructs the image from partial inpaintings, thus addressing the drawbacks of auto-enocoding methods. RIAD is extensively evaluated on several benchmarks and sets a new state-of-the art on a recent highly challenging anomaly detection benchmark.
ISSN:0031-3203
1873-5142
DOI:10.1016/j.patcog.2020.107706