Deep image generative modeling and statistical testing for industrial anomaly detection

This thesis addresses the challenge of anomaly detection in images, for industrial applications. It explores advanced methodologies employing both classical image processing techniques and modern generative modeling approaches, specifically focusing on Normalizing Flows and Diffusion Models. As anom...

Full description

Saved in:
Bibliographic Details
Main Author: Tailanián, Matías (author)
Format: doctoralThesis
Language:English
Published: 2024
Subjects:
Online Access:https://hdl.handle.net/20.500.12008/49470
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1868889936937091072
author Tailanián, Matías
author_browse Tailanián, Matías
author_facet Tailanián, Matías
author_role author
collection COLIBRI
dc.contributor.none.fl_str_mv Tailanián Matías, Universidad de la República (Uruguay). Facultad de Ingeniería.
dc.creator.none.fl_str_mv Tailanián, Matías
dc.date.none.fl_str_mv 2024
2025-04-02T15:00:19Z
2025-04-02T15:00:19Z
dc.format.none.fl_str_mv 206 p.
application/pdf
dc.identifier.none.fl_str_mv Tailanián, M. Deep image generative modeling and statistical testing for industrial anomaly detection [en línea]. Tesis de doctorado. Montevideo : Udelar. FI. IIE, 2024.
1688-2784
https://hdl.handle.net/20.500.12008/49470
dc.language.none.fl_str_mv en
eng
dc.publisher.none.fl_str_mv Udelar.FI
dc.rights.none.fl_str_mv info:eu-repo/semantics/openAccess
Licencia Creative Commons Atribución - No Comercial - Sin Derivadas (CC - By-NC-ND 4.0)
dc.source.none.fl_str_mv reponame:COLIBRI
instname:Universidad de la República
instacron:Universidad de la República
dc.subject.none.fl_str_mv Anomaly
Anomaly detection
Industrial anomaly detection
Image generative modeling
Diffusion models
Likelihood estimation
A contrario
NFA
Number of false alarms
Image processing
AI
Artificial intelligence
Machine learning
Anomalías
Detección de anomalías
Detección de anomalías industriales
Modelado de imágenes generativo
Modelos de difusión
Estimación de verosimilitud
NFA
Número de falsas alarmas
Procesamiento de imágenes
IA
Inteligencia artificial
Aprendizaje automático
dc.title.none.fl_str_mv Deep image generative modeling and statistical testing for industrial anomaly detection
dc.type.none.fl_str_mv Tesis de doctorado
info:eu-repo/semantics/doctoralThesis
info:eu-repo/semantics/acceptedVersion
description This thesis addresses the challenge of anomaly detection in images, for industrial applications. It explores advanced methodologies employing both classical image processing techniques and modern generative modeling approaches, specifically focusing on Normalizing Flows and Diffusion Models. As anomalies are rare by definition, collecting normal samples is generally easier and more feasible in industrial settings than acquiring comprehensive datasets with labeled anomalies. Therefore, the focus of this research is on unsupervised methods, and one-class methods, where the idea is to model the “normality” and detect anomalies as everything that deviates from this model. Initially, a multi-scale anomaly detection method based on classical image processing techniques is proposed, leveraging an a contrario approach to control the number of false alarms. Subsequently, a novel method called U-Flow is introduced, which employs a U-shaped architecture in Normalizing Flows to achieve anomaly detection with automatic thresholding. Then, this thesis further explores the use of Diffusion Models for anomaly detection, presenting the Diffusion Anomaly Detection (DAD) method. This work incorporates scorebased generative models and inpainting techniques to refine anomaly detection capabilities. Additionally, a new diffusion-based method called RIFA (Random Inpainting For Anomaly detection) is proposed as a completely unsupervised alternative. Finally, the techniques and knowledge gained from Diffusion Models are applied to a completely different application: counter-forensics. Throughout the whole thesis, a special emphasis is placed on bridging the gap between theoretical research and practical industrial applications, setting the theoretical foundations for obtaining automatic segmentations of anomalies, by performing statistical tests and controlling the number of false alarms using the a contrario framework. Experimental results on standard datasets validate the effectiveness of the proposed methods, highlighting substantial performance gains in some cases. The final chapter applies the best-performing method to two industrial problems : quality control in manufacturing leather samples for the upholstery industry, and defect detection in fruits, demonstrating its practical viability and impact on improving quality control processes in these industries In addition, this research contributes to the open-source community with several code repositories and has resulted in four published papers so far, and hopefully, more will follow. Future work will particularly focus on improving likelihood estimation with Diffusion Models and expanding its applicability to other industrial domains.
eu_rights_str_mv openAccess
format doctoralThesis
id anni_06495e28ca7d541a7877ccc736ff3d00
identifier_str_mv Tailanián, M. Deep image generative modeling and statistical testing for industrial anomaly detection [en línea]. Tesis de doctorado. Montevideo : Udelar. FI. IIE, 2024.
1688-2784
instacron_str Universidad de la República
institution Universidad de la República
instname_str Universidad de la República
language eng
language_invalid_str_mv en
network_acronym_str anni
network_name_str oai-lr-anni
oai_identifier_str oai:colibri.udelar.edu.uy:20.500.12008/49470
publishDate 2024
publishDateSort 2024
publisher.none.fl_str_mv Udelar.FI
reponame_str COLIBRI
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
rights_invalid_str_mv Licencia Creative Commons Atribución - No Comercial - Sin Derivadas (CC - By-NC-ND 4.0)
spelling Deep image generative modeling and statistical testing for industrial anomaly detectionTailanián, MatíasAnomalyAnomaly detectionIndustrial anomaly detectionImage generative modelingDiffusion modelsLikelihood estimationA contrarioNFANumber of false alarmsImage processingAIArtificial intelligenceMachine learningAnomalíasDetección de anomalíasDetección de anomalías industrialesModelado de imágenes generativoModelos de difusiónEstimación de verosimilitudNFANúmero de falsas alarmasProcesamiento de imágenesIAInteligencia artificialAprendizaje automáticoThis thesis addresses the challenge of anomaly detection in images, for industrial applications. It explores advanced methodologies employing both classical image processing techniques and modern generative modeling approaches, specifically focusing on Normalizing Flows and Diffusion Models. As anomalies are rare by definition, collecting normal samples is generally easier and more feasible in industrial settings than acquiring comprehensive datasets with labeled anomalies. Therefore, the focus of this research is on unsupervised methods, and one-class methods, where the idea is to model the “normality” and detect anomalies as everything that deviates from this model. Initially, a multi-scale anomaly detection method based on classical image processing techniques is proposed, leveraging an a contrario approach to control the number of false alarms. Subsequently, a novel method called U-Flow is introduced, which employs a U-shaped architecture in Normalizing Flows to achieve anomaly detection with automatic thresholding. Then, this thesis further explores the use of Diffusion Models for anomaly detection, presenting the Diffusion Anomaly Detection (DAD) method. This work incorporates scorebased generative models and inpainting techniques to refine anomaly detection capabilities. Additionally, a new diffusion-based method called RIFA (Random Inpainting For Anomaly detection) is proposed as a completely unsupervised alternative. Finally, the techniques and knowledge gained from Diffusion Models are applied to a completely different application: counter-forensics. Throughout the whole thesis, a special emphasis is placed on bridging the gap between theoretical research and practical industrial applications, setting the theoretical foundations for obtaining automatic segmentations of anomalies, by performing statistical tests and controlling the number of false alarms using the a contrario framework. Experimental results on standard datasets validate the effectiveness of the proposed methods, highlighting substantial performance gains in some cases. The final chapter applies the best-performing method to two industrial problems : quality control in manufacturing leather samples for the upholstery industry, and defect detection in fruits, demonstrating its practical viability and impact on improving quality control processes in these industries In addition, this research contributes to the open-source community with several code repositories and has resulted in four published papers so far, and hopefully, more will follow. Future work will particularly focus on improving likelihood estimation with Diffusion Models and expanding its applicability to other industrial domains.Subvención de financiación parcial de la Agencia Nacional de Investigación e Innovación de Uruguay.Udelar.FITailanián Matías, Universidad de la República (Uruguay). Facultad de Ingeniería.2025-04-02T15:00:19Z2025-04-02T15:00:19Z2024Tesis de doctoradoinfo:eu-repo/semantics/doctoralThesisinfo:eu-repo/semantics/acceptedVersion206 p.application/pdfTailanián, M. Deep image generative modeling and statistical testing for industrial anomaly detection [en línea]. Tesis de doctorado. Montevideo : Udelar. FI. IIE, 2024.1688-2784https://hdl.handle.net/20.500.12008/49470reponame:COLIBRIinstname:Universidad de la Repúblicainstacron:Universidad de la RepúblicaenengLas obras depositadas en el Repositorio se rigen por la Ordenanza de los Derechos de la Propiedad Intelectual de la Universidad de la República.(Res. Nº 91 de C.D.C. de 8/III/1994 – D.O. 7/IV/1994) y por la Ordenanza del Repositorio Abierto de la Universidad de la República (Res. Nº 16 de C.D.C. de 07/10/2014)info:eu-repo/semantics/openAccessLicencia Creative Commons Atribución - No Comercial - Sin Derivadas (CC - By-NC-ND 4.0)oai:colibri.udelar.edu.uy:20.500.12008/494702026-04-14T10:27:47Z
spellingShingle Deep image generative modeling and statistical testing for industrial anomaly detection
Tailanián, Matías
Anomaly
Anomaly detection
Industrial anomaly detection
Image generative modeling
Diffusion models
Likelihood estimation
A contrario
NFA
Number of false alarms
Image processing
AI
Artificial intelligence
Machine learning
Anomalías
Detección de anomalías
Detección de anomalías industriales
Modelado de imágenes generativo
Modelos de difusión
Estimación de verosimilitud
NFA
Número de falsas alarmas
Procesamiento de imágenes
IA
Inteligencia artificial
Aprendizaje automático
status_str acceptedVersion
title Deep image generative modeling and statistical testing for industrial anomaly detection
title_full Deep image generative modeling and statistical testing for industrial anomaly detection
title_fullStr Deep image generative modeling and statistical testing for industrial anomaly detection
title_full_unstemmed Deep image generative modeling and statistical testing for industrial anomaly detection
title_short Deep image generative modeling and statistical testing for industrial anomaly detection
title_sort Deep image generative modeling and statistical testing for industrial anomaly detection
topic Anomaly
Anomaly detection
Industrial anomaly detection
Image generative modeling
Diffusion models
Likelihood estimation
A contrario
NFA
Number of false alarms
Image processing
AI
Artificial intelligence
Machine learning
Anomalías
Detección de anomalías
Detección de anomalías industriales
Modelado de imágenes generativo
Modelos de difusión
Estimación de verosimilitud
NFA
Número de falsas alarmas
Procesamiento de imágenes
IA
Inteligencia artificial
Aprendizaje automático
url https://hdl.handle.net/20.500.12008/49470