PVIIAD: industrial visual anomaly detection method based on partially visible image inpainting

Abstract

Industrial anomaly detection is dedicated to identifying and locating regions that deviate from the standard appearance. The prevailing approach achieves unsupervised anomaly detection through the reconstruction of images using autoencoders. Due to the simplistic structure of some abnormal regions, the autoencoder can effectively reconstruct these areas, consequently diminishing the model’s anomaly detection capabilities. To address this issue, this paper transforms the reconstruction task into the inpainting-filling-reconstruction task to increase the reconstruction error between abnormal samples and normal samples. The masked regions inpainted by the filling network are used to fill in the input image, thereby achieving an effect similar to masking. Unlike typical masking processes, this approach retains partial authentic information in the input image, rendering it partially visible. This is beneficial for the reconstruction network to repair the masked areas. Due to the consistent structure between the masked region inpainted by the filling network and the normal region, the filled abnormal regions display a complex structure that has not been learned, making it difficult for the reconstruction network to reconstruct the abnormal regions. Experimental results indicate that our method performs better than other methods on both the MVTec AD dataset and the MVTec LOCO AD dataset.

Keywords

A

References

[1] Zheng X Q, Zheng S, Kong Y Get al.. Recent advances in surface defect inspection of industrial products using deep learning techniques. International journal of advanced manufacturing technology, 2021, 113(1–2): 35-58 J]

[2] Zhang R F, Wang H, Yang G P. MFFA: multi-level feature fusion and anomaly map compensation for anomaly detection. Journal of intelligent & fuzzy systems, 2023, 44(5): 7195-7210[J]

[3] Defard T, Setkov A, Loesch Aet al.. PaDiM: a patch distribution modeling framework for anomaly detection and localization. Proceedings of the 25th International Conference on Pattern Recognition Workshops, January 10–15, 2021, Virtual, 2021, Heidelberg, Springer Science and Business Media Deutschland GmbH475489[C]

[4] Tan M, Le Q V. EfficientNet: rethinking model scaling for convolutional neural networks. Proceedings of the 36th International Conference on Machine Learning, June 9–15, 2019, Long Beach, CA, USA, 2019, Washington, IMLS1069110700[C]

[5] Li C L, Sohn K, Yoon Jet al.. CutPaste: self-supervised learning for anomaly detection and localization. Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 19–25, 2021, Virtual, 2021, Los Alamitos, IEEE Computer Society96599669[C]

[6] Jiang J L, Zhu J L, Bilal Met al.. Masked swin transformer UNet for industrial anomaly detection. IEEE transactions on industrial informatics, 2023, 19(2): 2200-2209 J]

[7] Gong D, Liu L, Le Vet al.. Memorizing normality to detect anomaly: memory-augmented deep autoencoder for unsupervised anomaly detection. Proceedings of the 17th IEEE/CVF International Conference on Computer Vision, October 27–November 2, 2019, Seoul, Korea, 2019, New York, IEEE17051714[C]

[8] Xing P, Li Z C. Visual anomaly detection via partition memory bank module and error estimation. IEEE transactions on circuits and systems for video technology, 2023, 33(8): 3596-3607 J]

[9] Yang Z Y, Zhang T, Bozchalooi I Set al.. Memory-augmented generative adversarial networks for anomaly detection. IEEE transactions on neural networks and learning systems, 2022, 33(6): 2324-2334 J]

[10] Liang Y F, Zhang J N, Zhao S Wet al.. Omni-frequency channel-selection representations for unsupervised anomaly detection. IEEE transactions on image processing, 2023, 32: 4327-4340 J]

[11] ZAVRTANIK V, KRISTAN M, SKOCAJ D. Reconstruction by inpainting for visual anomaly detection[J]. Pattern recognition, 2021, 112.

[12] Wang Z, Bovik A C, Sheikh H Ret al.. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 2004, 13(4): 600-612 J]

[13] Zhang B, Sander P V, Bermak A. Gradient magnitude similarity deviation on multiple scales for color image quality assessment. Proceedings of the 2017 IEEE International Conference on Acoustics, Speech, and Signal Processing, March 5–9, 2017, New Orleans, LA, USA, 2017, New York, IEEE12531257[C]

[14] Bergmann P, Batzner K, Fauser Met al.. The MVTec anomaly detection dataset: a comprehensive real-world dataset for unsupervised anomaly detection. International journal of computer vision, 2021, 129(4): 1038-1059 J]

[15] Bergmann P, Batzner K, Fauser Met al.. Beyond dents and scratches: logical constraints in unsupervised anomaly detection and localization. International journal of computer vision, 2022, 130(4): 947-969 J]

[16] Wu R Z, Li S, Chen C Let al.. Improving video anomaly detection performance by mining useful data from unseen video frames. Neurocomputing, 2021, 462: 523-533 J]

[17] CAI X Y, XIAO R L, ZENG Z X, et al. ITran: a novel transformer-based approach for industrial anomaly detection and localization[J]. Engineering applications of artificial intelligence, 2023, 125.

[18] Zhou Q, He S, Liu Het al.. Pull&Push: leveraging differential knowledge distillation for efficient unsupervised anomaly detection and localization. IEEE transactions on circuits and systems for video technology, 2023, 33(5): 2176-2189 J]

[19] Liu T K, Li B, Du Xet al.. Component-aware anomaly detection framework for adjustable and logical industrial visual inspection. Advanced engineering informatics, 2023, 58: 14 J]

[20] NIV C, ISSAR T, YEDID H. Set features for fine-grained anomaly detection[EB/OL]. (2023-02-23) [2024-03-27]. https://arxiv.org/abs/2302.12245.

RIGHTS & PERMISSIONS

Tianjin University of Technology