Search

Supplemental Material, Supplemental_Tables_4,5,6_Profiles_of_seizures_ver_12-27-19_at_1039hrs for Risk Factors for Seizures and Epilepsy in Children With Congenital Heart Disease by Suman Ghosh, Joseph Philip, Nikita Patel, Jennifer Munoz-Pareja, Dalia Lopez-Colon, Mark Bleiweis and Steven Parrish Winesett in Journal of Child Neurology

El text es la manera més habitual d’intercanviar informació des de les edats. Amb el desenvolupament recent d’imatges de bases de dades de manuscrits manuscrits històrics, la demanda d’algorismes per fer accessibles aquestes bases de dades per a la navegació i la indexació estan augmentant. L’augment exponencial de les bases de dades d’imatges disponibles públicament i les col·leccions personals d’imatges, aquest interès també inclou l’enteniment del text sobre imatges naturals. Activar la cerca o comprendre una gran collecció de manuscrits o bases de dades d’imatges requereix un mètode ràpid i robust. Els investigadors han trobat diferents maneres de representar paraules retallades per a la comprensió i la concordança, cosa que funciona bé quan les paraules ja estan segmentades. No obstant això, no hi ha cap manera trivial d’estendre’ls per a documents no segmentats. En aquesta tesi, explorem diferents mètodes per a la recuperació i el reconeixement de text a partir d’imatges sense escena de documents i escenes. Hi ha dues formes de representació diferents a la literatura, s’utilitza una representació de longitud fixa a partir de paraules retallades i una altra amb una seqüència de característiques de longitud variable. Al llarg d’aquesta tesi, hem estudiat aquestes dues representacions per la seva idoneïtat en la lliure comprensió del text. A la primera part, ens hem centrat en la segmentació de paraules lliures amb una representació de longitud fixa. Vam ampliar l’ús d’una representació exitosa per a la recuperació gratuïta de segmentacions. En la segona part de la tesi, explorem funcions basades en seqüències i, finalment, proposem una solució unificada on el mateix marc pot generar ambdós tipus de representacions. El texto es la forma más común de compartir información desde edades. Con el reciente desarrollo de bases de datos de imágenes de manuscritos históricos manuscritos, la demanda de algoritmos para hacer accesibles estas bases de datos para la navegación y la indexación están en aumento. El aumento exponencial de las bases de datos de imágenes disponibles al público y las colecciones personales de imágenes, este interés ahora también abarca la comprensión de texto en imágenes naturales. La habilitación de la búsqueda o la comprensión de una gran colección de manuscritos o bases de datos de imágenes necesita un método rápido y sólido. Los investigadores han encontrado diferentes maneras de representar la palabra recortada para la comprensi ón y la coincidencia, que funciona bien cuando las palabras ya están segmentadas. Sin embargo, no hay una forma trivial de ampliar estos para documentos no segmentados. En esta tesis, exploramos diferentes métodos para la recuperación y el reconocimiento de texto a partir de documentos no segmentados e imégenes de escena. Existen dos formas diferentes de representación en la literatura, una usa una representaci ón de longitud fija aprendida de palabras recortadas y otra una secuencia de características de longitud variable. A lo largo de esta tesis, hemos estudiado estas dos representaciones por su idoneidad en la segmentación, la comprensión libre del texto. En la primera parte, nos centramos en la detección de palabras sin segmentación utilizando una representación de longitud fija. Extendimos el uso de una representación exitosa para la recuperación libre de segmentación. En la segunda parte de la tesis, exploramos las características basadas en secuencia y finalmente proponemos una soluci´on unificada donde el mismo marco puede generar ambos tipos de representaciones. Text is the most common way of information sharing from ages. With recent development of images databases of handwritten historic manuscripts the demand for algorithms to make these databases accessible for browsing and indexing are in rise. Exponential increase of publicly available image databases and personal collections of pictures, this interest now also embraces text understanding on natural images. Enabling search or understanding large collection of manuscripts or image databases need fast and robust method. Researchers have found different ways to represent cropped word for understanding and matching, which works well in when words are already segmented. However there is no trivial way to extend these for non-segmented documents. In this thesis we explore different methods for text retrieval and recognition from unsegmented document and scene images. Two different ways of representation exists in literature, one uses a fixed length representation learned from cropped words and another a sequence of features of variable length. Throughout this thesis, we have studied both these representation for their suitability in segmentation free understanding of text. In the first part we are focused on segmentation free word spotting using a fixed length representation. We extended the use of one successful representation for segmentation free retrieval. In the second part of the thesis, we explore sequence based features and finally propose a unified solution where same framework can generate both kind of representations.

Offline signature verification is one of the most challenging tasks in biometrics and document forensics. Unlike other verification problems, it needs to model minute but critical details between genuine and forged signatures, because a skilled falsification might often resembles the real signature with small deformation. This verification task is even harder in writer independent scenarios which is undeniably fiscal for realistic cases. In this paper, we model an offline writer independent signature verification task with a convolutional Siamese network. Siamese networks are twin networks with shared weights, which can be trained to learn a feature space where similar observations are placed in proximity. This is achieved by exposing the network to a pair of similar and dissimilar observations and minimizing the Euclidean distance between similar pairs while simultaneously maximizing it between dissimilar pairs. Experiments conducted on cross-domain datasets emphasize the capability of our network to model forgery in different languages (scripts) and handwriting styles. Moreover, our designed Siamese network, named SigNet, exceeds the state-of-the-art results on most of the benchmark signature datasets, which paves the way for further research in this direction.

This paper proposes a region based convolutional neural network for segmentation-free word spotting. Our net- work takes as input an image and a set of word candidate bound- ing boxes and embeds all bounding boxes into an embedding space, where word spotting can be casted as a simple nearest neighbour search between the query representation and each of the candidate bounding boxes. We make use of PHOC embedding as it has previously achieved significant success in segmentation- based word spotting. Word candidates are generated using a simple procedure based on grouping connected components using some spatial constraints. Experiments show that R-PHOC which operates on images directly can improve the current state-of- the-art in the standard GW dataset and performs as good as PHOCNET in some cases designed for segmentation based word spotting., Accepted in ICDAR'2017

This study focused on how and why secondary school mathematics teachers might situate real-world equity issues in their lessons. Real-world equity issues in the context of mathematics education are real-world issues which might be critically examined in the mathematics lesson and so encourage pupils to be democratic citizens who are critically literate through mathematics. The first phase of the study involved data collection through semi-structured interviews with eight teachers to learn about teachers’ mathematical beliefs. Four of the participants had mathematics degrees and the others had degrees in other disciplines. An adapted version of Ernest’s model of mathematics-related belief systems was used as card sort prompts for the interview. The second phase of the study collected data from observing the participants teach a mathematics lesson in which they situated real-world equity issues. It also gathered data on the participants’ reflections of the lesson. The data formed the basis of eight case studies. In order to identify the different ways in which real-world equity issues can be situated in the mathematics classroom, data from the lesson observations and teachers’ reflections were used to address the question: ‘How might secondary school mathematics teachers situate real-world equity issues in their lessons?’. The data was analysed by identifying the areas of the curriculum addressed in the lesson, and Skovsmose’s Milieus of Learning matrix was used as a framework to analyse the structure of the lesson. Data from interviews and teachers’ reflections were used to address the question: ‘Why might secondary school mathematics teachers situate real-world equity issues in their lessons?’. Ernest’s model of mathematics-related belief systems was used to analyse the interview data and identify what motivated teachers to situate real-world equity issues in their lessons in the way they did. As the cohort comprised of mathematics teachers with mathematics degrees and degrees from other disciplines, the study also analysed the data between participants to determine if there is a difference between teachers from diverse academic backgrounds in terms of their mathematical beliefs and practices. By drawing on the analysis, the study arrived at conclusions to provide potential ways in which teachers from diverse beliefs and academic backgrounds might be able to situate real-world equity issues in the mathematics classroom.