Supervision : Matthieu CORD
Co-supervision : GANÇARSKI Stéphane
Distance metric learning for image and webpage comparison
This thesis focuses on distance metric learning for image and webpage comparison. Distance metrics are used in many machine learning and computer vision contexts such as k-nearest neighbors classification, clustering, support vector machine, information/image retrieval, visualization etc. In this thesis, we focus on Mahalanobis-like distance metric learning where the learned model is parametered by a symmetric positive semidefinite matrix. It learns a linear tranformation such that the Euclidean distance in the induced projected space satisfies learning constraints.
First, we propose a method based on comparison between relative distances that takes rich relations between data into account, and exploits similarities between quadruplets of examples. We apply this method on relative attributes and hierarchical image classification.
Second, we propose a new regularization method that controls the rank of the learned matrix, limiting the number of independent parameters and overfitting. We show the interest of our method on synthetic and real-world recognition datasets.
Eventually, we propose a novel Webpage change detection framework in a context of archiving. For this purpose, we use temporal distance relations between different versions of a same Webpage. The metric learned in a totally unsupervised way detects important regions and ignores unimportant content such as menus and advertisements. We show the interest of our method on different Websites.
Defence : 01/20/2015 - 10h30
Jury members :
Patrick Pérez, Technicolor (Rennes), Rapporteur
Alain Rakotomamonjy, Université de Rouen (Rouen), Rapporteur
Francis Bach, Inria - Ecole Normale Supérieure (Paris), Examinateur
Patrick Gallinari, UPMC (Paris), Examinateur
Jean Ponce, Ecole Normale Supérieure (Paris), Examinateur
Frédéric Précioso, Polytech'Nice-Sophia, Examinateur
Matthieu Cord, UPMC (Paris), Directeur de thèse
Stéphane Gançarski, UPMC (Paris), Co-directeur de thèse
Nicolas Thome, UPMC (Paris), Invité
- Marc T. Law, N. Thome, M. Cord : “Learning a Distance Metric from Relative Comparisons between Quadruplets of Images”, International Journal of Computer Vision, pp. 1-30, (Springer Verlag) (2016)
- M. Law : “Apprentissage de distance pour la comparaison d’images et de pages Web”, thesis, defence 01/20/2015, supervision Cord, Matthieu, rapporteurs : GANÇARSKI Stéphane (2015)
- M. Law, N. Thome, S. Gançarski, M. Cord : “Apprentissage de métrique appliqué à la détection de changement de page Web et aux attributs relatifs”, CORIA 2015 - Conférence en Recherche d'Infomations et Applications - 12th French Information Retrieval Conference, Paris, France (2015)
- M. Law, N. Thome, M. Cord : “Fantope Regularization in Metric Learning”, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Columbus, Ohio, United States, pp. 1051-1058 (2014)
- M. Law, N. Thome, M. Cord : “Bag-of-Words Image Representation: Key Ideas and Further Insight”, chapter in Fusion in Computer Vision - Understanding Complex Visual Content, Advances in Computer Vision and Pattern Recognition, pp. 29-52, (Springer) (2014)
- M. Law, N. Thome, M. Cord : “Quadruplet-Wise Image Similarity Learning”, IEEE International Conference on Computer Vision (ICCV), Sydney, Australia, pp. 249-256 (2013)
- M. Law, N. Thome, M. Cord : “Hybrid Pooling Fusion in the BoW Pipeline”, ECCV 2012 Workshop on Information fusion in Computer Vision for Concept Recognition (ECCV-IFCVCR 2012), vol. 7585, Lecture Notes in Computer Science, Florence, Italy, pp. 355-364, (Springer) (2012)
- M. Law, N. Thome, S. Gançarski, M. Cord : “Structural and Visual Comparisons for Web Page Archiving”, 12th edition of the ACM Symposium on Document Engineering, DocEng'12, Paris, France, pp. 117-120, (ACM) (2012)
- M. Law, C. Sureda Gutierrez, N. Thome, S. Gançarski, M. Cord : “Structural and Visual Similarity Learning for Web Page Archiving”, 10th workshop on Content-Based Multimedia Indexing (CBMI), Annecy, France, pp. 1-6, (IEEE) (2012)