Soutenance de thèse Maurits Diephuis

M. Maurits Diephuis soutiendra, en vue de l'obtention du grade de docteur ès sciences, mention informatique, sa thèse intitulée:

Micro-structure Based Physical Object Identification on Mobile Platforms


Sous la direction de:

  • Prof. S. Voloshynovskiy, Département d'informatique, Faculté des sciences, Université de Genève


Physical object protection includes all techniques to identify or authenticate objects to determine their origin. High quality counterfeited products are ever more prevalent and widespread nowadays. Luxurious items have always been popular targets, but the last two decades, driven by ever cheaper and sophisticated manufacturing technology, have also seen fake medication, fake industrial and aerospace parts. 

Popular countermeasure techniques tend to invasively alter an object, for example by adding markings, holograms, chips or using expensive printing techniques. These techniques hinge on the assumption that they are either to hard or to expensive to replicate. 

In this thesis we envision a different approach to physical object protection based on the microscopic surface structure of the object’s surface. This micro-structure is both unique to the object and currently non-cloneable and thus serves as natural identifier. More- over, micro-structure based security schemes have relatively cheap enrollment, are non- invasive leaving the original object untouched. Lastly, verification be done by ordinary consumers without any particular expertise.

This thesis will extend the state-of-the-art in several important ways. It show cases a number of methods to allow micro-structure based object protection on hand-held mobile platforms, both for enrollment and verification without any kind of modification of the acquisition device or lighting. Secondly, the developed algorithms only require the object to be in the field of view and do not need any aid, not in the form of a mark on the package, or software on the mobile phone, to acquire and extract the sought micro-structure.

The process that enables this is not trivial. Optically acquired micro-structures are visually poor, lacking edges or salient regions. Further degradation is caused by the mobile phone, the angle under which it is held, and under what lighting. Further-more, while one can expect the number of enrolled objects to be huge, there are only but a few acquisitions available per unique sample, to learn from.

This thesis proposes a number of solutions for micro-structure verification and frame- works in which they can be applied, explicitly modeling user selectable parameters and environmental factors that influence the design and performance of micro-structure based authentication and identification frameworks.

Firstly, this thesis demonstrates how micro-structures may be matched based on so called robust features and geometry. Specifically, it proposes an affinity based algorithm that can match small sets of points, of which over 50% are outliers.

Secondly, a novel robust descriptor was developed and is patent pending: Sketchprint. Specifically designed to robustly identify distorted micro structures, it requires no train- ing, is relatively stable and information rich, and requires but a small number of enrolled descriptor vectors per sample. As it both captures geometrical and micro-structure information from its region of interest, it doesn’t require any exhaustive geometrical re-ranking, nor aggregation.

Tangent to Sketchprint, to address aggregation and compact descriptor representations, a statistical model of a Bag-of-Word content identification model has been built and theoretically analyzed. It captures all relevant parameters, from the type and quantity of the used descriptors, the desired robustness to noise, architectural choices such as the deployed type of pooling, and ties those in to the predicted end performance. This model has also been empirically verified.

A propos de Maurits:


Date: Lundi 29 mai 2017 à 13h30

Lieu: Battelle bâtiment A - Auditoire rez-de-chaussée

24 mai 2017

À la Une