On-Line Handwritten Character Recognition using Kohonen Networks

This paper presents an efficient Online Handwritten character Recognition System for Malayalam Characters (OHR-M) using Kohonen network. It would help in recognizing Malayalam text entered using pen-like devices. It will be more natural and efficient way for users to enter text using a pen than keyboard and mouse. To identify the difference between similar characters in Malayalam a novel feature extraction method has been adopted-a combination of context bitmap and normalized (x, y) coordinates. The system reported an accuracy of 88.75% which is writer independent with a recognition time of 15-32 milliseconds

HelF, a suppressor of RNAi mediated gene silencing in Dictyostelium discoideum

RNA interference (RNAi) is a recently discovered process, in which double stranded RNA (dsRNA) triggers the homology-dependant degradation of cognate messenger RNA (mRNA). In a search for new components of the RNAi machinery in Dictyostelium, a new gene was identified, which was called helF. HelF is a putative RNA helicase, which shows a high homology to the helicase domain of Dicer, to the helicase domain of Dictyostelium RdRP and to the C. elegans gene drh-1, that codes for a dicer related DExH-box RNA helicase, which is required for RNAi. The aim of the present Ph.D. work was to investigate the role of HelF in PTGS, either induced by RNAi or asRNA. A genomic disruption of the helF gene was performed, which resulted in a distinct mutant morphology in late development. The cellular localization of the protein was elucidated by creating a HelF-GFP fusion protein, which was found to be localized in speckles in the nucleus. The involvement of HelF in the RNAi mechanism was studied. For this purpose, RNAi was induced by transformation of RNAi hairpin constructs against four endogenous genes in wild type and HelF- cells. The silencing efficiency was strongly enhanced in the HelF K.O. strain in comparison with the wild type. One gene, which could not be silenced in the wild type background, was successfully silenced in HelF-. When the helF gene was disrupted in a secondary transformation in a non-silenced strain, the silencing efficiency was strongly improved, a phenomenon named here “retrosilencing”. Transcriptional run-on experiments revealed that the enhanced gene silencing in HelF- was a posttranscriptional event, and that the silencing efficiency depended on the transcription levels of hairpin RNAs. In HelF-, the threshold level of hairpin transcription required for efficient silencing was dramatically lowered. The RNAi-mediated silencing was accompanied by the production of siRNAs; however, their amount did not depend on the level of hairpin transcription. These results indicated that HelF is a natural suppressor of RNAi in Dictyostelium. In contrast, asRNA mediated gene silencing was not enhanced in the HelF K.O, as shown for three tested genes. These results confirmed previous observations (H. Martens and W. Nellen, unpublished) that although similar, RNAi and asRNA mediated gene silencing mechanisms differ in their requirements for specific proteins. In order to characterize the function of the HelF protein on a molecular level and to study its interactions with other RNAi components, in vitro experiments were performed. Besides the DEAH-helicase domain, HelF contains a double-stranded RNA binding domain (dsRBD) at its N-terminus, which showed high similarity to the dsRBD domain of Dicer A from Dictyostelium. The ability of the recombinant dsRBDs from HelF and Dicer A to bind dsRNA was examined and compared. It was shown by gel-shift assays that both HelF-dsRBD and Dicer-dsRBD could bind directly to long dsRNAs. However, HelF-dsRBD bound more efficiently to dsRNA with imperfect matches than to perfect dsRNA. Both dsRBDs bound specifically to a pre-miRNA substrate (pre-let-7). The results suggested that most probably there were two binding sites for the proteins on the pre-miRNA substrate. Moreover, it was shown that HelF-dsRBD and Dicer-dsRBD have siRNA-binding activity. The affinities of the two dsRBDs to the pre-let-7 substrate were also examined by plasmon surface resonance analyses, which revealed a 9-fold higher binding affinity of the Dicer-dsRBD to pre-let-7 compared to that of the HelF-dsRBD. The binding of HelF-dsRBD to the pre-let-7 was impaired in the presence of Mg2+, while the Dicer-dsRBD interaction with pre-let-7 was not influenced by the presence of Mg2+. The results obtained in this thesis can be used to postulate a model for HelF function. In this, HelF acts as a nuclear suppressor of RNAi in wild type cells by recognition and binding of dsRNA substrates. The protein might act as a surveillance system to avoid RNAi initiation by fortuitous dsRNA formation or low abundance of dsRNA trigger. If the protein acts as an RNA helicase, it could unwind fold-back structures in the nucleus and thus lead to decreased RNAi efficiency. A knock-out of HelF would result in initiation of the RNAi pathway even by low levels of dsRNA. The exact molecular function of the protein in the RNAi mechanism still has to be elucidated. RNA interferenz (RNAi) ist ein in jüngster Zeit entdeckter Mechanismus, bei dem doppelsträngige RNA Moleküle (dsRNA) eine Homologie-abhängige Degradation einer verwandten messenger-RNA (mRNA) auslösen. Auf der Suche nach neuen Komponenten der RNAi-Maschinerie in Dictyostelium konnte ein neues Gen (helF) identifiziert werden. HelF ist eine putative RNA-Helikase mit einer hohen Homologie zur Helikasedomäne der bekannten Dicerproteine, der Helikasedomäne der Dictyostelium RdRP und zu dem C. elegans Gen drh-1, welches für eine Dicer-bezogene DExH-box RNA Helikase codiert, die am RNAi-Mechanismus beteiligt ist. Das Ziel dieser Arbeit war es, die Funktion von HelF im Zusammenhang des RNAi oder asRNA induzierten PTGS zu untersuchen. Es wurde eine Unterbrechung des helF-Gens auf genomischer Ebene (K.O.) vorgenommen, was bei den Mutanten zu einer veränderten Morphologie in der späten Entwicklung führte. Die Lokalisation des Proteins in der Zelle konnte mit Hilfe einer GFP-Fusion analysiert werden und kleinen Bereichen innerhalb des Nukleus zugewiesen werden. Im Weiteren wurde der Einfluss von HelF auf den RNAi-Mechanismus untersucht. Zu diesem Zweck wurde RNAi durch Einbringen von RNAi Hairpin-Konstrukten gegen vier endogene Gene im Wiltypstamm und der HelF--Mutante induziert. Im Vergleich zum Wildtypstamm konnte im HelF--Mutantenstamm eine stark erhöhte „Silencing“-Effizienz nachgewiesen werden. Ein Gen, welches nach RNAi Initiation im Wildtypstamm unverändert blieb, konnte im HelF--Mutantenstamm erfolgreich stillgelegt werden. Durch sekundäres Einführen einer Gendisruption im helF-Locus in einen Stamm, in welchem ein Gen nicht stillgelegt werden konnte, wurde die Effizienz des Stilllegens deutlich erhöht. Dieses Phänomen wurde hier erstmals als „Retrosilencing“ beschrieben. Mit Hilfe von transkriptionellen run-on Experimenten konnte belegt werden, dass es sich bei dieser erhöhten Stilllegungseffizienz um ein posttranskriptionelles Ereignis handelte, wobei die Stillegungseffizienz von der Transkriptionsstärke der Hairpin RNAs abhängt. Für die HelF--Mutanten konnte gezeigt werden, dass der Schwellenwert zum Auslösen eines effizienten Stillegens dramatisch abgesenkt war. Obwohl die RNAi-vermittelte Genstilllegung immer mit der Produktion von siRNAs einhergeht, war die Menge der siRNAs nicht abhängig von dem Expressionsniveau des Hairpin-Konstruktes. Diese Ergebnisse legen nahe, dass es sich bei der HelF um einen natürlichen Suppressor des RNAi-Mechanismus in Dictyostelium handelt. Im Gegensatz hierzu war die as-vermittelte Stilllegung von drei untersuchten Genen im HelF-K.O. im Vergleich zum Wildyp unverändert. Diese Ergebnisse bestätigten frühere Beobachtungen (H. Martens und W. Nellen, unveröffentlicht), wonach die Mechanismen für RNAi und asRNA-vermittelte Genstilllegung unterschiedliche spezifische Proteine benötigen. Um die Funktion des HelF-Proteins auf der molekularen Ebene genauer zu charakterisieren und die Interaktion mit anderen RNAi-Komponenten zu untersuchen, wurden in vitro Versuche durchgeführt. Das HelF-Protein enthält, neben der DEAH-Helikase-Domäne eine N-terminale Doppelstrang RNA bindende Domäne (dsRBD) mit einer hohen Ähnlichkeit zu der dsRBD des Dicer A aus Dictyostelium. Die dsRNA-Bindungsaktivität der beiden dsRBDs aus HelF und Dicer A wurde analysiert und verglichen. Es konnte mithilfe von Gel-Retardationsanalysen gezeigt werden, dass sowohl HelF-dsRBD als auch Dicer-dsRBD direkt an lange dsRNAs binden können. Hierbei zeigte sich, dass die HelF-dsRBD eine höhere Affinität zu einem imperfekten RNA-Doppelstrang besitzt, als zu einer perfekt gepaarten dsRNA. Für beide dsRBDs konnte eine spezifische Bindung an ein pre-miRNA Substrat nachgewiesen werden (pre-let-7). Dieses Ergebnis legt nah, dass es zwei Bindestellen für die Proteine auf dem pre-miRNA Substrat gibt. Überdies hinaus konnte gezeigt werden, dass die dsRBDs beider Proteine eine siRNA bindende Aktivität besitzen. Die Affinität beider dsRBDs an das pre-let-7 Substrat wurde weiterhin mit Hilfe der Plasmon Oberflächen Resonanz untersucht. Hierbei konnte eine 9-fach höhere Bindeaffinität der Dicer-dsRBD im Vergleich zur HelF-dsRBD nachgewiesen werden. Während die Bindung der HelF-dsRBD an das pre-let-7 durch die Anwesenheit von Mg2+ beeinträchtigt war, zeigte sich kein Einfluß von Mg2+ auf das Bindeverhalten der Dicer-dsRBD. Mit Hilfe der in dieser Arbeit gewonnen Ergebnisse lässt sich ein Model für die Funktion von HelF postulieren. In diesem Model wirkt HelF durch Erkennen und Binden von dsRNA Substraten als Suppressor von der RNAi im Kern. Das Protein kann als Überwachungsystem gegen eine irrtümliche Auslösung von RNAi wirken, die durch zufällige dsRNA Faltungen oder eine zu geringe Häufigkeit der siRNAs hervorgerufen sein könnte. Falls das Protein eine Helikase-Aktivität besitzt, könnte es rückgefaltete RNA Strukturen im Kern auflösen, was sich in einer verringerten RNAi-Effizienz wiederspiegelt. Durch Ausschalten des helF-Gens würde nach diesem Modell eine erfolgreiche Auslösung von RNAi schon bei sehr geringer Mengen an dsRNA möglich werden. Das Modell erlaubt, die exakte molekulare Funktion des HelF-Proteins im RNAi-Mechanismus weiter zu untersuchen.

Characterization of DnmA, the Dnmt2 homolog in Dictyostelium discoideum

Eukaryotic DNA m5C methyltransferases (MTases) play a major role in many epigenetic regulatory processes like genomic imprinting, X-chromosome inactivation, silencing of transposons and gene expression. Members of the two DNA m5C MTase families, Dnmt1 and Dnmt3, are relatively well studied and many details of their biological functions, biochemical properties as well as interaction partners are known. In contrast, the biological functions of the highly conserved Dnmt2 family, which appear to have non-canonical dual substrate specificity, remain enigmatic despite the efforts of many researchers. The genome of the social amoeba Dictyostelium encodes Dnmt2-homolog, the DnmA, as the only DNA m5C MTase which allowed us to study Dnmt2 function in this organism without interference by the other enzymes. The dnmA gene can be easily disrupted but the knock-out clones did not show obvious phenotypes under normal lab conditions, suggesting that the function of DnmA is not vital for the organism. It appears that the dnmA gene has a low expression profile during vegetative growth and is only 5-fold upregulated during development. Fluorescence microscopy indicated that DnmA-GFP fusions were distributed between both the nucleus and cytoplasm with some enrichment in nuclei. Interestingly, the experiments showed specific dynamics of DnmA-GFP distribution during the cell cycle. The proteins colocalized with DNA in the interphase and were mainly removed from nuclei during mitosis. DnmA functions as an active DNA m5C MTase in vivo and is responsible for weak but detectable DNA methylation of several regions in the Dictyostelium genome. Nevertheless, gel retardation assays showed only slightly higher affinity of the enzyme to dsDNA compared to ssDNA and no specificity towards various sequence contexts, although weak but detectable specificity towards AT-rich sequences was observed. This could be due to intrinsic curvature of such sequences. Furthermore, DnmA did not show denaturant-resistant covalent complexes with dsDNA in vitro, although it could form covalent adducts with ssDNA. Low binding and methyltransfer activity in vitro suggest the necessity of additional factor in DnmA function. Nevertheless, no candidates could be identified in affinity purification experiments with different tagged DnmA fusions. In this respect, it should be noted that tagged DnmA fusion preparations from Dictyostelium showed somewhat higher activity in both covalent adduct formation and methylation assays than DnmA expressed in E.coli. Thus, the presence of co-purified factors cannot be excluded. The low efficiency of complex formation by the recombinant enzyme and the failure to define interacting proteins that could be required for DNA methylation in vivo, brought up the assumption that post-translational modifications could influence target recognition and enzymatic activity. Indeed, sites of phosphorylation, methylation and acetylation were identified within the target recognition domain (TRD) of DnmA by mass spectrometry. For phosphorylation, the combination of MS data and bioinformatic analysis revealed that some of the sites could well be targets for specific kinases in vivo. Preliminary 3D modeling of DnmA protein based on homology with hDNMT2 allowed us to show that several identified phosphorylation sites located on the surface of the molecule, where they would be available for kinases. The presence of modifications almost solely within the TRD domain of DnmA could potentially modulate the mode of its interaction with the target nucleic acids. DnmA was able to form denaturant-resistant covalent intermediates with several Dictyostelium tRNAs, using as a target C38 in the anticodon loop. The formation of complexes not always correlated with the data from methylation assays, and seemed to be dependent on both sequence and structure of the tRNA substrate. The pattern, previously suggested by the Helm group for optimal methyltransferase activity of hDNMT2, appeared to contribute significantly in the formation of covalent adducts but was not the only feature of the substrate required for DnmA and hDNMT2 functions. Both enzymes required Mg2+ to form covalent complexes, which indicated that the specific structure of the target tRNA was indispensable. The dynamics of covalent adduct accumulation was different for DnmA and different tRNAs. Interestingly, the profiles of covalent adduct accumulation for different tRNAs were somewhat similar for DnmA and hDNMT2 enzymes. According to the proposed catalytic mechanism for DNA m5C MTases, the observed denaturant-resistant complexes corresponded to covalent enamine intermediates. The apparent discrepancies in the data from covalent complex formation and methylation assays may be interpreted by the possibility of alternative pathways of the catalytic mechanism, leading not to methylation but to exchange or demethylation reactions. The reversibility of enamine intermediate formation should also be considered. Curiously, native gel retardation assays showed no or little difference in binding affinities of DnmA to different RNA substrates and thus the absence of specificity in the initial enzyme binding. The meaning of the tRNA methylation as well as identification of novel RNA substrates in vivo should be the aim of further experiments.

Hierarchical Object Recognition Using Libraries of Parameterized Model Sub-Parts

This thesis describes the development of a model-based vision system that exploits hierarchies of both object structure and object scale. The focus of the research is to use these hierarchies to achieve robust recognition based on effective organization and indexing schemes for model libraries. The goal of the system is to recognize parameterized instances of non-rigid model objects contained in a large knowledge base despite the presence of noise and occlusion. Robustness is achieved by developing a system that can recognize viewed objects that are scaled or mirror-image instances of the known models or that contain components sub-parts with different relative scaling, rotation, or translation than in models. The approach taken in this thesis is to develop an object shape representation that incorporates a component sub-part hierarchy- to allow for efficient and correct indexing into an automatically generated model library as well as for relative parameterization among sub-parts, and a scale hierarchy- to allow for a general to specific recognition procedure. After analysis of the issues and inherent tradeoffs in the recognition process, a system is implemented using a representation based on significant contour curvature changes and a recognition engine based on geometric constraints of feature properties. Examples of the system's performance are given, followed by an analysis of the results. In conclusion, the system's benefits and limitations are presented.

The Use of Grouping in Visual Object Recognition.

The report describes a recognition system called GROPER, which performs grouping by using distance and relative orientation constraints that estimate the likelihood of different edges in an image coming from the same object. The thesis presents both a theoretical analysis of the grouping problem and a practical implementation of a grouping system. GROPER also uses an indexing module to allow it to make use of knowledge of different objects, any of which might appear in an image. We test GROPER by comparing it to a similar recognition system that does not use grouping.

Statistical Object Recognition

Two formulations of model-based object recognition are described. MAP Model Matching evaluates joint hypotheses of match and pose, while Posterior Marginal Pose Estimation evaluates the pose only. Local search in pose space is carried out with the Expectation--Maximization (EM) algorithm. Recognition experiments are described where the EM algorithm is used to refine and evaluate pose hypotheses in 2D and 3D. Initial hypotheses for the 2D experiments were generated by a simple indexing method: Angle Pair Indexing. The Linear Combination of Views method of Ullman and Basri is employed as the projection model in the 3D experiments.

Geometry and Photometry in 3D Visual Recognition

The report addresses the problem of visual recognition under two sources of variability: geometric and photometric. The geometric deals with the relation between 3D objects and their views under orthographic and perspective projection. The photometric deals with the relation between 3D matte objects and their images under changing illumination conditions. Taken together, an alignment-based method is presented for recognizing objects viewed from arbitrary viewing positions and illuminated by arbitrary settings of light sources.

Attentional Selection in Object Recognition

A key problem in object recognition is selection, namely, the problem of identifying regions in an image within which to start the recognition process, ideally by isolating regions that are likely to come from a single object. Such a selection mechanism has been found to be crucial in reducing the combinatorial search involved in the matching stage of object recognition. Even though selection is of help in recognition, it has largely remained unsolved because of the difficulty in isolating regions belonging to objects under complex imaging conditions involving occlusions, changing illumination, and object appearances. This thesis presents a novel approach to the selection problem by proposing a computational model of visual attentional selection as a paradigm for selection in recognition. In particular, it proposes two modes of attentional selection, namely, attracted and pay attention modes as being appropriate for data and model-driven selection in recognition. An implementation of this model has led to new ways of extracting color, texture and line group information in images, and their subsequent use in isolating areas of the scene likely to contain the model object. Among the specific results in this thesis are: a method of specifying color by perceptual color categories for fast color region segmentation and color-based localization of objects, and a result showing that the recognition of texture patterns on model objects is possible under changes in orientation and occlusions without detailed segmentation. The thesis also presents an evaluation of the proposed model by integrating with a 3D from 2D object recognition system and recording the improvement in performance. These results indicate that attentional selection can significantly overcome the computational bottleneck in object recognition, both due to a reduction in the number of features, and due to a reduction in the number of matches during recognition using the information derived during selection. Finally, these studies have revealed a surprising use of selection, namely, in the partial solution of the pose of a 3D object.

An Analysis of the Effect of Gaussian Error in Object Recognition

Object recognition is complicated by clutter, occlusion, and sensor error. Since pose hypotheses are based on image feature locations, these effects can lead to false negatives and positives. In a typical recognition algorithm, pose hypotheses are tested against the image, and a score is assigned to each hypothesis. We use a statistical model to determine the score distribution associated with correct and incorrect pose hypotheses, and use binary hypothesis testing techniques to distinguish between them. Using this approach we can compare algorithms and noise models, and automatically choose values for internal system thresholds to minimize the probability of making a mistake.

Object Recognition with Pictorial Structures

This thesis presents a statistical framework for object recognition. The framework is motivated by the pictorial structure models introduced by Fischler and Elschlager nearly 30 years ago. The basic idea is to model an object by a collection of parts arranged in a deformable configuration. The appearance of each part is modeled separately, and the deformable configuration is represented by spring-like connections between pairs of parts. These models allow for qualitative descriptions of visual appearance, and are suitable for generic recognition problems. The problem of detecting an object in an image and the problem of learning an object model using training examples are naturally formulated under a statistical approach. We present efficient algorithms to solve these problems in our framework. We demonstrate our techniques by training models to represent faces and human bodies. The models are then used to locate the corresponding objects in novel images.

Surface Reflectance Recognition and Real-World Illumination Statistics

Humans distinguish materials such as metal, plastic, and paper effortlessly at a glance. Traditional computer vision systems cannot solve this problem at all. Recognizing surface reflectance properties from a single photograph is difficult because the observed image depends heavily on the amount of light incident from every direction. A mirrored sphere, for example, produces a different image in every environment. To make matters worse, two surfaces with different reflectance properties could produce identical images. The mirrored sphere simply reflects its surroundings, so in the right artificial setting, it could mimic the appearance of a matte ping-pong ball. Yet, humans possess an intuitive sense of what materials typically "look like" in the real world. This thesis develops computational algorithms with a similar ability to recognize reflectance properties from photographs under unknown, real-world illumination conditions. Real-world illumination is complex, with light typically incident on a surface from every direction. We find, however, that real-world illumination patterns are not arbitrary. They exhibit highly predictable spatial structure, which we describe largely in the wavelet domain. Although they differ in several respects from the typical photographs, illumination patterns share much of the regularity described in the natural image statistics literature. These properties of real-world illumination lead to predictable image statistics for a surface with given reflectance properties. We construct a system that classifies a surface according to its reflectance from a single photograph under unknown illuminination. Our algorithm learns relationships between surface reflectance and certain statistics computed from the observed image. Like the human visual system, we solve the otherwise underconstrained inverse problem of reflectance estimation by taking advantage of the statistical regularity of illumination. For surfaces with homogeneous reflectance properties and known geometry, our system rivals human performance.

Rotation Invariant Real-time Face Detection and Recognition System

In this report, a face recognition system that is capable of detecting and recognizing frontal and rotated faces was developed. Two face recognition methods focusing on the aspect of pose invariance are presented and evaluated - the whole face approach and the component-based approach. The main challenge of this project is to develop a system that is able to identify faces under different viewing angles in realtime. The development of such a system will enhance the capability and robustness of current face recognition technology. The whole-face approach recognizes faces by classifying a single feature vector consisting of the gray values of the whole face image. The component-based approach first locates the facial components and extracts them. These components are normalized and combined into a single feature vector for classification. The Support Vector Machine (SVM) is used as the classifier for both approaches. Extensive tests with respect to the robustness against pose changes are performed on a database that includes faces rotated up to about 40 degrees in depth. The component-based approach clearly outperforms the whole-face approach on all tests. Although this approach isproven to be more reliable, it is still too slow for real-time applications. That is the reason why a real-time face recognition system using the whole-face approach is implemented to recognize people in color video sequences.

A Detailed Look at Scale and Translation Invariance in a Hierarchical Neural Model of Visual Object Recognition

The HMAX model has recently been proposed by Riesenhuber & Poggio as a hierarchical model of position- and size-invariant object recognition in visual cortex. It has also turned out to model successfully a number of other properties of the ventral visual stream (the visual pathway thought to be crucial for object recognition in cortex), and particularly of (view-tuned) neurons in macaque inferotemporal cortex, the brain area at the top of the ventral stream. The original modeling study only used ``paperclip'' stimuli, as in the corresponding physiology experiment, and did not explore systematically how model units' invariance properties depended on model parameters. In this study, we aimed at a deeper understanding of the inner workings of HMAX and its performance for various parameter settings and ``natural'' stimulus classes. We examined HMAX responses for different stimulus sizes and positions systematically and found a dependence of model units' responses on stimulus position for which a quantitative description is offered. Interestingly, we find that scale invariance properties of hierarchical neural models are not independent of stimulus class, as opposed to translation invariance, even though both are affine transformations within the image plane.