Learning Embeddings for Indexing, Retrieval, and Classification, with Applications to Object and Shape Recognition in Image Databases

Nearest neighbor retrieval is the task of identifying, given a database of objects and a query object, the objects in the database that are the most similar to the query. Retrieving nearest neighbors is a necessary component of many practical applications, in fields as diverse as computer vision, pattern recognition, multimedia databases, bioinformatics, and computer networks. At the same time, finding nearest neighbors accurately and efficiently can be challenging, especially when the database contains a large number of objects, and when the underlying distance measure is computationally expensive. This thesis proposes new methods for improving the efficiency and accuracy of nearest neighbor retrieval and classification in spaces with computationally expensive distance measures. The proposed methods are domain-independent, and can be applied in arbitrary spaces, including non-Euclidean and non-metric spaces. In this thesis particular emphasis is given to computer vision applications related to object and shape recognition, where expensive non-Euclidean distance measures are often needed to achieve high accuracy. The first contribution of this thesis is the BoostMap algorithm for embedding arbitrary spaces into a vector space with a computationally efficient distance measure. Using this approach, an approximate set of nearest neighbors can be retrieved efficiently - often orders of magnitude faster than retrieval using the exact distance measure in the original space. The BoostMap algorithm has two key distinguishing features with respect to existing embedding methods. First, embedding construction explicitly maximizes the amount of nearest neighbor information preserved by the embedding. Second, embedding construction is treated as a machine learning problem, in contrast to existing methods that are based on geometric considerations. The second contribution is a method for constructing query-sensitive distance measures for the purposes of nearest neighbor retrieval and classification. In high-dimensional spaces, query-sensitive distance measures allow for automatic selection of the dimensions that are the most informative for each specific query object. It is shown theoretically and experimentally that query-sensitivity increases the modeling power of embeddings, allowing embeddings to capture a larger amount of the nearest neighbor structure of the original space. The third contribution is a method for speeding up nearest neighbor classification by combining multiple embedding-based nearest neighbor classifiers in a cascade. In a cascade, computationally efficient classifiers are used to quickly classify easy cases, and classifiers that are more computationally expensive and also more accurate are only applied to objects that are harder to classify. An interesting property of the proposed cascade method is that, under certain conditions, classification time actually decreases as the size of the database increases, a behavior that is in stark contrast to the behavior of typical nearest neighbor classification systems. The proposed methods are evaluated experimentally in several different applications: hand shape recognition, off-line character recognition, online character recognition, and efficient retrieval of time series. In all datasets, the proposed methods lead to significant improvements in accuracy and efficiency compared to existing state-of-the-art methods. In some datasets, the general-purpose methods introduced in this thesis even outperform domain-specific methods that have been custom-designed for such datasets.

Use of semiconductor optical amplifiers in signal processing applications

We describe a 42.6 Gbit/s all-optical pattern recognition system which uses semiconductor optical amplifiers (SOAs). A circuit with three SOA-based logic gates is used to identify the presence of specific port numbers in an optical packet header.

Electronically controlled surface plasmon dispersion and optical transmission through metallic hole arrays using liquid crystal.

The enhanced optical properties of metal films periodically perforated with an array of sub-wavelength size holes have recently been widely studied in the field of surface plasmon optics. The ability to design the optical transmission of such nanostructures, which act as plasmonic crystals, by varying their geometrical parameters gives them great flexibility for numerous applications in photonics, opto-electronics, and sensing. Transforming these passive optical elements into devices that may be actively controlled has presented a new challenge. Here, we report on the realization of an electrically controlled nanostructured optical system based on the unique properties of surface plasmon polaritonic crystals in contact with a liquid crystal (LC) layer. We discuss the effect of LC layer modulation on the surface plasmon dispersion, the related optical transmission and the underlying mechanism. The reported effect may be used to achieve active spectral tuneability and switching in a wide range of applications.

The tautomerism, solvatochromism and non-linear optical properties of fluorescent 3-hydroxyquinoxaline-2-carboxalidine-4-aminoantipyrine

The Schiff base, 3-hydroxyquinoxaline-2-carboxalidine-4-aminoantipyrine, was synthesized by the condensation of 3-hydroxyquinoxaline-2-carboxaldehyde with 4-aminoantipyrine. HPLC, FT-IR and NMR spectral data revealed that the compound exists predominantly in the amide tautomeric form and exhibits both absorption and fluorescence solvatochromism, large stokes shift, two electron quasireversible redox behaviour and good thermal stability, with a glass transition temperature of 104oC. The third-order non-linear optical character was studied using open aperture Z-scan methodology employing 7 ns pulses at 532 nm. The third-order non-linear absorption coefficient, b, was 1.48 x 10-6 cm W-1 and the imaginary part of the third-order non-linear optical susceptibility, Im c(3), was 3.36 x10-10 esu. The optical limiting threshold for the compound was found to be 340 MW cm-2.

The tautomerism, solvatochromism and non-linear optical properties of fluorescent 3-hydroxyquinoxaline-2-carboxalidine-4-aminoantipyrine

The Schiff base, 3-hydroxyquinoxaline-2-carboxalidine-4-aminoantipyrine, was synthesized by the condensation of 3-hydroxyquinoxaline-2-carboxaldehyde with 4-aminoantipyrine. HPLC, FT-IR and NMR spectral data revealed that the compound exists predominantly in the amide tautomeric form and exhibits both absorption and fluorescence solvatochromism, large stokes shift, two electron quasireversible redox behaviour and good thermal stability, with a glass transition temperature of 104 oC. The third-order non-linear optical character was studied using open aperture Z-scan methodology employing 7 ns pulses at 532 nm. The third-order non-linear absorption coefficient, b, was 1.48 x 10-6 cm W-1 and the imaginary part of the third-order non-linear optical susceptibility, Im c(3), was 3.36x10-10 esu. The optical limiting threshold for the compound was found to be 340 MW cm-2.

A System for Offline Recognition of Handwritten Characters in Malayalam Script

In this paper, we propose a handwritten character recognition system for Malayalam language. The feature extraction phase consists of gradient and curvature calculation and dimensionality reduction using Principal Component Analysis. Directional information from the arc tangent of gradient is used as gradient feature. Strength of gradient in curvature direction is used as the curvature feature. The proposed system uses a combination of gradient and curvature feature in reduced dimension as the feature vector. For classification, discriminative power of Support Vector Machine (SVM) is evaluated. The results reveal that SVM with Radial Basis Function (RBF) kernel yield the best performance with 96.28% and 97.96% of accuracy in two different datasets. This is the highest accuracy ever reported on these datasets

Subpixel-Filterung für eine autostereoskopische Multiperspektiven-3-D-Darstellung hoher Qualität

Die stereoskopische 3-D-Darstellung beruht auf der naturgetreuen Präsentation verschiedener Perspektiven für das rechte und linke Auge. Sie erlangt in der Medizin, der Architektur, im Design sowie bei Computerspielen und im Kino, zukünftig möglicherweise auch im Fernsehen, eine immer größere Bedeutung. 3-D-Displays dienen der zusätzlichen Wiedergabe der räumlichen Tiefe und lassen sich grob in die vier Gruppen Stereoskope und Head-mounted-Displays, Brillensysteme, autostereoskopische Displays sowie echte 3-D-Displays einteilen. Darunter besitzt der autostereoskopische Ansatz ohne Brillen, bei dem N≥2 Perspektiven genutzt werden, ein hohes Potenzial. Die beste Qualität in dieser Gruppe kann mit der Methode der Integral Photography, die sowohl horizontale als auch vertikale Parallaxe kodiert, erreicht werden. Allerdings ist das Verfahren sehr aufwendig und wird deshalb wenig genutzt. Den besten Kompromiss zwischen Leistung und Preis bieten präzise gefertigte Linsenrasterscheiben (LRS), die hinsichtlich Lichtausbeute und optischen Eigenschaften den bereits früher bekannten Barrieremasken überlegen sind. Insbesondere für die ergonomisch günstige Multiperspektiven-3-D-Darstellung wird eine hohe physikalische Monitorauflösung benötigt. Diese ist bei modernen TFT-Displays schon recht hoch. Eine weitere Verbesserung mit dem theoretischen Faktor drei erreicht man durch gezielte Ansteuerung der einzelnen, nebeneinander angeordneten Subpixel in den Farben Rot, Grün und Blau. Ermöglicht wird dies durch die um etwa eine Größenordnung geringere Farbauflösung des menschlichen visuellen Systems im Vergleich zur Helligkeitsauflösung. Somit gelingt die Implementierung einer Subpixel-Filterung, welche entsprechend den physiologischen Gegebenheiten mit dem in Luminanz und Chrominanz trennenden YUV-Farbmodell arbeitet. Weiterhin erweist sich eine Schrägstellung der Linsen im Verhältnis von 1:6 als günstig. Farbstörungen werden minimiert, und die Schärfe der Bilder wird durch eine weniger systematische Vergrößerung der technologisch unvermeidbaren Trennelemente zwischen den Subpixeln erhöht. Der Grad der Schrägstellung ist frei wählbar. In diesem Sinne ist die Filterung als adaptiv an den Neigungswinkel zu verstehen, obwohl dieser Wert für einen konkreten 3-D-Monitor eine Invariante darstellt. Die zu maximierende Zielgröße ist der Parameter Perspektiven-Pixel als Produkt aus Anzahl der Perspektiven N und der effektiven Auflösung pro Perspektive. Der Idealfall einer Verdreifachung wird praktisch nicht erreicht. Messungen mit Hilfe von Testbildern sowie Schrifterkennungstests lieferten einen Wert von knapp über 2. Dies ist trotzdem als eine signifikante Verbesserung der Qualität der 3-D-Darstellung anzusehen. In der Zukunft sind weitere Verbesserungen hinsichtlich der Zielgröße durch Nutzung neuer, feiner als TFT auflösender Technologien wie LCoS oder OLED zu erwarten. Eine Kombination mit der vorgeschlagenen Filtermethode wird natürlich weiterhin möglich und ggf. auch sinnvoll sein.

Use of decision trees in colour feature selection: application to object recognition in outdoor scenes

A new method for the automated selection of colour features is described. The algorithm consists of two stages of processing. In the first, a complete set of colour features is calculated for every object of interest in an image. In the second stage, each object is mapped into several n-dimensional feature spaces in order to select the feature set with the smallest variables able to discriminate the remaining objects. The evaluation of the discrimination power for each concrete subset of features is performed by means of decision trees composed of linear discrimination functions. This method can provide valuable help in outdoor scene analysis where no colour space has been demonstrated as being the most suitable. Experiment results recognizing objects in outdoor scenes are reported

A new approach to real time traffic sign recognition based on colour information

The automatic interpretation of conventional traffic signs is very complex and time consuming. The paper concerns an automatic warning system for driving assistance. It does not interpret the standard traffic signs on the roadside; the proposal is to incorporate into the existing signs another type of traffic sign whose information will be more easily interpreted by a processor. The type of information to be added is profuse and therefore the most important object is the robustness of the system. The basic proposal of this new philosophy is that the co-pilot system for automatic warning and driving assistance can interpret with greater ease the information contained in the new sign, whilst the human driver only has to interpret the "classic" sign. One of the codings that has been tested with good results and which seems to us easy to implement is that which has a rectangular shape and 4 vertical bars of different colours. The size of these signs is equivalent to the size of the conventional signs (approximately 0.4 m2). The colour information from the sign can be easily interpreted by the proposed processor and the interpretation is much easier and quicker than the information shown by the pictographs of the classic signs

Efficient Coxian duration modelling for activity recognition in smart environment with the hidden semi-Markov model

In this paper, we exploit the discrete Coxian distribution and propose a novel form of stochastic model, termed as the Coxian hidden semi-Makov model (Cox-HSMM), and apply it to the task of recognising activities of daily living (ADLs) in a smart house environment. The use of the Coxian has several advantages over traditional parameterization (e.g. multinomial or continuous distributions) including the low number of free parameters needed, its computational efficiency, and the existing of closed-form solution. To further enrich the model in real-world applications, we also address the problem of handling missing observation for the proposed Cox-HSMM. In the domain of ADLs, we emphasize the importance of the duration information and model it via the Cox-HSMM. Our experimental results have shown the superiority of the Cox-HSMM in all cases when compared with the standard HMM. Our results have further shown that outstanding recognition accuracy can be achieved with relatively low number of phases required in the Coxian, thus making the Cox-HSMM particularly suitable in recognizing ADLs whose movement trajectories are typically very long in nature.

Modeling Reflective Bistability in Vertical-Cavity Semiconductor Optical Amplifiers

The characteristics of optical bistability in a vertical- cavity semiconductor optical amplifier (VCSOA) operated in reflection are reported. The dependences of the optical bistability in VCSOAs on the initial phase detuning and on the applied bias current are analyzed. The optical bistability is also studied for different numbers of superimposed periods in the top distributed bragg reflector (DBR) that conform the internal cavity of the device. The appearance of the X-bistable and the clockwise bistable loops is predicted theoretically in a VCSOA operated in reflection for the first time, to the best of our knowledge. Moreover, it is also predicted that the control of the VCSOA’s top reflectivity by the addition of new superimposed periods in its top DBR reduces by one order of magnitude the input power needed for the assessment of the X- and the clockwise bistable loop, compared to that required in in-plane semiconductor optical amplifiers. These results, added to the ease of fabricating two-dimensional arrays of this kind of device could be useful for the development of new optical logic or optical signal regeneration devices.

Recognition of online handwritten music symbols

Paper submitted to MML 2013, 6th International Workshop on Machine Learning and Music, Prague, September 23, 2013.

The Illinois pattern recognition computer (Illiac III) /

"Presented at the 18th Annual ACM National Conference, Denver, Colorado August 27, 1963"

Efficient use of training data in the n-tuple recognition method

A simple technique is presented for improving the robustness of the n-tuple recognition method against inauspicious choices of architectural parameters, guarding against the saturation problem, and improving the utilisation of small data sets. Experiments are reported which confirm that the method significantly improves performance and reduces saturation in character recognition problems.