Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.

Miglioramento della qualità di algoritmi di face morphing generation

L'obiettivo principale di questo lavoro di tesi è quello di migliorare gli algoritmi di morphing generation in termini di qualità visiva e di potenzialità di attacco dei sistemi automatici di riconoscimento facciale.

Sperimentazione di metodologie innovative di indagine termografica in campo industriale

La termografia è un metodo d’indagine ampiamente utilizzato nei test diagnostici non distruttivi, in quanto risulta una tecnica d’indagine completamente non invasiva, ripetibile nel tempo e applicabile in diversi settori. Attraverso tale tecnica è possibile individuare difetti superficiali e sub–superficiali, o possibili anomalie, mediante la rappresentazione della distribuzione superficiale di temperatura dell’oggetto o dell’impianto indagato. Vengono presentati i risultati di una campagna sperimentale di rilevamenti termici, volta a stabilire i miglioramenti introdotti da tecniche innovative di acquisizione termografica, quali ad esempio la super-risoluzione, valutando un caso di studio di tipo industriale. Si è effettuato un confronto tra gli scatti registrati, per riuscire a individuare e apprezzare le eventuali differenze tra le diverse modalità di acquisizione adottate. L’analisi dei risultati mostra inoltre come l’utilizzo dei dispositivi di acquisizione termografica in modalità super-resolution sia possibile anche su cavalletto.

Progettazione e implementazione di un sistema di autorizzazione elettronico per il trasporto auto di persone disabili con validazione biometrica

Notizie riguardanti scandali relativi al utilizzo inappropriato di contrassegni per disabili sono all’ordine del giorno. Situazioni meno popolari dal punto di vista mediatico, ma altrettanto gravi a livello sociale coinvolgono tutti quegli individui che si prodigano a falsificare contrassegni oppure ad utilizzarli anche in mancanza del disabile, eventualmente anche successivamente al decesso del medesimo. Tutto questo va inevitabilmente a discapito di tutti coloro che hanno reale diritto e necessità di usufruire delle agevolazioni. Lo scopo di questa tesi è quindi quello di illustrare un possibile sistema per contrastare e possibilmente debellare questo malcostume diffusissimo in Italia. La proposta è quella di dematerializzare il pass cartaceo sostituendolo con un equiva- lente elettronico, temporaneo e associato non più ad una targa, ma all’individuo stesso. Per farlo si ricorrerà all’uso di tecniche di autenticazione attraverso sistemi biometrici, quali il riconoscimento facciale, vocale, di espressioni facciali e gestures.

Tecniche di riconoscimento del volto in presenza di makeup

Il makeup, come strumento atto a modificare i tratti somatici di un individuo per aumentarne la bellezza, è largamente diffuso e socialmente accettabile al giorno d’oggi. Per la sua facile reperibilità, semplicità di utilizzo e capacità di alterare le caratteristiche principali di un volto, può diventare uno strumento pericoloso per chi volesse sottrarsi a dei controlli. In questo lavoro di tesi sono stati analizzati algoritmi presenti in letteratura che cercano di arginare gli effetti di alterazione di un viso, causati dal makeup, durante un processo di riconoscimento del volto. Inoltre è stato utilizzato un software per verificare la robustezza dei programmi commerciali in merito al problema del makeup e confrontare poi i risultati riscontrati in letteratura con quelli ottenuti dai test.

Smart Mirror: remote verification of an identity document holder

Nowadays, some activities, such as subscribing an insurance policy or opening a bank account, are possible by navigating through a web page or a downloadable application. Since the user is often “hidden” behind a monitor or a smartphone, it is necessary a solution able to guarantee about their identity. Companies are often requiring the submission of a “proof-of-identity”, which usually consists in a picture of an identity document of the user, together with a picture or a brief video of themselves. This work describes a system whose purpose is the automation of these kinds of verifications.

Generative Neural Networks for image iper-resolution and improvement of Optical Character Recognition's performances

The usage of Optical Character Recognition’s (OCR, systems is a widely spread technology into the world of Computer Vision and Machine Learning. It is a topic that interest many field, for example the automotive, where becomes a specialized task known as License Plate Recognition, useful for many application from the automation of toll road to intelligent payments. However, OCR systems need to be very accurate and generalizable in order to be able to extract the text of license plates under high variable conditions, from the type of camera used for acquisition to light changes. Such variables compromise the quality of digitalized real scenes causing the presence of noise and degradation of various type, which can be minimized with the application of modern approaches for image iper resolution and noise reduction. Oneclass of them is known as Generative Neural Networks, which are very strong ally for the solution of this popular problem.

Towards the kinetic resolution of mechanically planar chiral rotaxanes

In this work, we reported the synthesis and characterization of two [2]rotaxanes endowed with a central ammonium group and two triazolium recognition stations on either side, acting as complexation sites for a dibenzo-24-crown-8 ether macrocycle. These mechanically interlocked architectures were obtained through the interlocking of a functionalized achiral macrocycle with Cs symmetry (where the symmetry element is a mirror plane corresponding to plane of the ring) and a C∞v symmetric axle (where a mirror plane and a C∞ principal axis are aligned along the axle length). We took advantage of the reversible acid/base triggered molecular shuttling of the ring between two lateral triazolium units to switch the rotaxanes between prochiral and mechanically planar chiral forms, which exists as two rapidly-interconverting co-conformers. We exploited the reactivity of the central amino group to attach an optically pure chiral substituent, with the goal of demonstrating the enantiomeric nature of the co-conformers and to obtain a non-zero diastereomeric excess in the resulting diastereomeric products through a dynamic kinetic resolution. To this end, two enantiopure reagents were chosen that could perform clean and fast reaction with amines: a sulfonyl chloride and an acyl chloride. Only the acyl chloride successfully produced an amide in high yield with the deprotonated rotaxane. The group added to the central amine station acted as a stopper against the shuttling of the macrocycle along the axis, thus preventing the fast interconversion of the two mechanically planar enantiomers. We analysed the results through static and dynamic NMR spectroscopic techniques by varying temperature and solvent used. Indeed, the presence of diastereomers was recorded alongside the configurational isomers resulting from the slow rotation of the CN-CO bond of the amide moiety, thus paving the way for a dynamic kinetic resolution.