Detection theory and industrial applications
R. GROMPONE
Image processingModelling

Objectif du cours

Our team at the Borelli Center is regularly confronted with industrial problems whose ultimate goal is the detection of objects or events in images or image series. The topic of classification and deep learning is widely discussed in several MVA courses. However, classification is not the same as detection, since detection is often characterized by the lack of a sufficient number of examples to reduce it to a classification problem.

This course introduces one of the theories that addresses the detection problem, namely the a-contrario theory. This statistical framework aims at automatically selecting detection thresholds to control the expected number of false detections under given background models.

The course will cover the theory and practice of image and video processing required for this type of automatic detection by analyzing several examples, including real-world applications.

Organisation des séances

The course will be held online and will consist of weekly lectures of 3 hours each, covering theory and discussion of particular cases. Students will be expected to submit regular exercises clarifying technical points of the course, as well as a final report analyzing a paper from the literature.

Mode de validation

Validation is based on 4 homework assignments, including exercises and a report analyzing a research paper.

Références

– Lecture notes.
– A Desolneux, L Moisan, JM Morel, « From gestalt theory to image analysis: a probabilistic approach », Springer, 2007.
– G Kanizsa, « Organization in Vision: Essays on Gestalt Perception », Praeger Publishers, 1979.
– A Gordon, G Glazko, X Qiu, A Yakovlev, « Control of the mean number of false discoveries, Bonferroni and stability of multiple testing », The Annals of Applied Statistics 1 (1), 179-190, 2007.
– Various research papers

Thèmes abordés

– Introduction: visual perception, Gestalt theory, discussion of human vision.
– A-contrario theory: concept of background model, number of false alarms (NFA), modeling the background and the structure to be detected.
– Detection of specific geometric structures.
– Time series detection in satellite images.
– Detection of image forgery.

Les intervenants

Rafael GROMPONE

voir les autres cours du 2nd semestre
Rechercher