Am Institut für Print- und Medientechnik an der TU Chemnitz wird ein System entwickelt, welches gescannte Dokumente archivieren soll. Im Gegensatz zu bereits existierenden OCR-Systemen, sollen diese Dokumente hier jedoch nicht mittels einer Schrifterkennung verarbeitet werden. Vielmehr sind Textbereiche in den gescannten Vorlagen zu vektorisieren. Bereiche mit Grafiken und Illustrationen werden bei diesem Verfahren als ein Bildvektor gespeichert. Diese Vorgehensweise soll es ermöglichen, auch Dokumente mit Schriftsymbolen effizient zu archivieren, die keinen "herkömmlichen" Schriftsätzen zugehörig sind. Diese Studienarbeit stellt Merkmalsextraktionsverfahren vor, die aus einem gegebenen Teil (Segment) eines Dokumentenscans Merkmale extrahieren, die es ermöglichen sollen, diesen mittels eines Klassifikationsverfahrens einer Klasse Textblock oder einer Klasse Grafikblock zuzuordnen. Zusätzlich werden zwei Klassifikationsverfahren, ein Entscheidungsbaum und eine Fuzzy-Logik, auf die Nutzbarkeit für einen solchen Mustererkennungsprozess überprüft. Als Textblöcke erkannte Bereiche werden im zu entwickelnden Gesamtverfahren dann in nachfolgenden Verarbeitungsschritten einer Vektorisierung zugeführt.

A way to follow products through the chain of production is important in the process industry and it is often solved by marking them with serial numbers. In some cases permanent markings such as dot peen marking is required. To ensure profitability in the industry and reduce errors, these markings must be read automatically. Automatic reading of dot peen markings using a camera can be hard since there is low contrast between the background and the numbers, the background can be uneven and different illuminations can affect the visibility. In this work, two different systems are implemented and evaluated to assess the possibility of developing a robust system. One system uses image analysis to segment the numbers before classifying them. The other system uses the recent advances in deep learning for object detection. Both implementations are shown to work in near real-time on a cpu. The deep learning object detection approach was able to classify all numbers correct in a image 60% of the time, while the other approach only succeeded in 20% of the time.

abstract: The Arizona State University Herbarium began in 1896 when Professor Fredrick Irish collected the first recorded Arizona specimen for what was then called the Tempe Normal School - a Parkinsonia microphylla. Since then, the collection has grown to approximately 400,000 specimens of vascular plants and lichens. The most recent project includes the digitization - both the imaging and databasing - of approximately 55,000 vascular plant specimens from Latin America. To accomplish this efficiently, possibilities in non-traditional methods, including both new and existing technologies, were explored. SALIX (semi-automatic label information extraction) was developed as the central tool to handle automatic parsing, along with BarcodeRenamer (BCR) to automate image file renaming by barcode. These two developments, combined with existing technologies, make up the SALIX Method. The SALIX Method provides a way to digitize herbarium specimens more efficiently than the traditional approach of entering data solely through keystroking. Using digital imaging, optical character recognition, and automatic parsing, I found that the SALIX Method processes data at an average rate that is 30% faster than typing. Data entry speed is dependent on user proficiency, label quality, and to a lesser degree, label length. This method is used to capture full specimen records, including close-up images where applicable. Access to biodiversity data is limited by the time and resources required to digitize, but I have found that it is possible to do so at a rate that is faster than typing. Finally, I experiment with the use of digital field guides in advancing access to biodiversity data, to stimulate public engagement in natural history collections. / Dissertation/Thesis / M.S. Plant Biology 2012

This paper presents an algorithm with the purpose of automatically dividing a simple floor plan into sections. Sections include names, size and location on the image, all of which will be automatically extracted by the algorithm as a step of converting a simple image into an interactive map. The labels for each section utilizes tesseract-OCR wrapper tesseractJS to extract text and label location. In regards to section borders pixel traversing is employed coupled with CIE76 for color comparison which results in the discovery of size and location of the section. Performance of the algorithm was measured on three different maps using metrics such as correctness, quality, completeness, jaccard index and name accuracy. The metrics showed the potential of such an algorithm in terms of automating the task of sectioning an image. With results ranging between lowest percentage of 48% and highest of 100% on three different maps looking at correctness, quality, completeness, average jaccard index and average name accuracy per map.

Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2016-01-18T12:53:32Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Dissertação_Pedro Gonçalves.pdf: 8565820 bytes, checksum: f14c8d37ea1f5daef226a8a6f131bada (MD5) / Made available in DSpace on 2016-01-18T12:53:32Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Dissertação_Pedro Gonçalves.pdf: 8565820 bytes, checksum: f14c8d37ea1f5daef226a8a6f131bada (MD5) Previous issue date: 2015-09-01 / Punir infrações de trânsito, controlar tráfego em rodovias, controlar o acesso a áreas restritas, entre outras, são ações tomadas para melhorar o trânsito nas grandes cidades. Para realizar tais ações é necessário, portanto, identificar o veículo automotivo, utilizando, para isso, sua placa de licenciamento. Entretanto, com o aumento de automóveis nas vias urbanas, essa tarefa tornou-se muito difícil de ser realizada de uma forma eficiente por apenas agentes de trânsito, pois existe uma grande quantidade de dados a serem analisados e reportados aos órgãos competentes. Soma-se a isso, o fato de fatores emocionais, cansaços físico e mental, inerentes aos seres humanos, atrapalharem a eficácia da tarefa executada. Por isso, ferramentas que realizam o reconhecimento ótico de caracteres, Opitcal Character Recognition (OCR), vem sendo cada vez mais empregadas para realizar a identificação automática de caracteres existentes nas placas dos automóveis. Este trabalho visa descrever um sistema para identificação de veículos automotivos através de imagens estáticas, apresentando técnicas pesquisadas e estudadas em cada etapa do processo de identificação. As etapas que são apresentadas e detalhadas incluem: a identificação da placa, segmentação dos caracteres presentes na placa e o reconhecimento dos caracteres isolados. Técnicas envolvendo processamento digital de imagem como detectores de bordas, operações morfológicas, análise de componentes conectados e limiarização serão explicitadas. Redes neurais artificias são propostas para realizar o reconhecimento do caractere isolado, tais como Self-Organizing Maps (SOM) e Kernel Self-Organizing Map (KSOM), e serão pormenorizadas. Para avaliar o desempenho das técnicas empregadas nesse projeto, imagens presentes na base de dados MediaLab LPR Database foram utilizadas. Métricas como Recall, Precision e F-Score foram empregadas na avaliação de performance dos diferentes algoritmos estudados e implementados para realizar a detecção da placa, ajudando na escolha do extrator da placa do sistema final. No estágio de segmentação da placa e do reconhecimento dos caracteres isolados, a taxa de acerto foi utilizada para avaliar os algoritmos propostos. Para um grupo de 276 imagens pertencentes a uma base pública, as etapas de detecção, segmentação e reconhecimento alcançaram desempenhos semelhantes aos vigentes na literatura ANAGNOSTOPOULOS et al. (2006) e propiciaram, aproximadamente, uma taxa de acerto global do sistema OCR proposto de 85%. / Punish traffic infractions, traffic control on highways, control access to restricted areas, among others, are actions taken to improve traffic in major cities. In order to take these actions is therefore necessary to identify the motor vehicle using it licensing plate. However, with the increase of the number of cars on urban roads, this task has become very difficult to be performed effectively only by traffic agents because there is a lot of data to be analyzed and reported to the competent agencies. In addition, the fact that emotional factors, physical and mental tiredness, that inherent features to humans, hider effectiveness of task begin performed. Therefore, tools that perform optical character recognition (OCR) are begin increasingly used for automating the identification of characters on licensing plate of the vehicles. This research describes a system for identification of automotive vehicles through still images showing algorithms researched in the literature on each step of the identification process. The stages are presented and detailed include: plate identification, segmentation of the characters existing in plate and the recognition of single characters. Techniques involving digital image processing like edge detectors, morphological operations, connected component analysis and thresholding are explained. Artificial neural networks are submitted to achieve the recognition of single character, such as Self-Organizing Maps (SOM) and Kernel Self-Organizing Map (KSOM), are detailed. In order to evaluate the performance of the techniques used in this project, images from mainly the MediaLab LPR Database were used. The metrics employed to analyze the performance of algorithms implemented to detect a region of plate on image are Recall, Precision and F-Score. This metrics helped to choose the better algorithms for extraction plate on image. In the segmentation stage of the plate and the recognition of single characters, the hit rate was used to evaluate the proposed algorithms. For group of 276 images belonging a public database, the stages of detection, segmentation and recognition reached similar performance with previous approaches (ANAGNOSTOPOULOS et al., 2006), leading the proposed OCR system to 85% of hit rate.

Nowadays, most of the documents are stored in electronic form and there is a high demand to organize and categorize them efficiently. Therefore, the field of automated text classification has gained a significant attention both from science and industry. This technology has been applied to information retrieval, information filtering, news classification, etc. The goal of this project is the automated text classification of photos as invoices or receipts in Visma Mobile Scanner, based on the previously extracted text. Firstly, several OCR tools available on the market have been evaluated in order to find the most accurate to be used for the text extraction, which turned out to be ABBYY FineReader. The machine learning tool WEKA has been used for the text classification, with the focus on the Naïve Bayes classifier. Since the Naïve Bayes implementation provided by WEKA does not support some advances in the text classification field such as N-gram, Laplace smoothing, etc., an improved version of Naïve Bayes classifier which is more specialized for the text classification and the invoice/receipt classification has been implemented. Improving the Naive Bayes classifier, investigating how it can be improved for the problem domain and evaluating the obtained classification accuracy compared to the generic Naïve Bayes are the main parts of this research. Experimental results show that the specialized Naïve Bayes classifier has the highest accuracy. By applying the Fixed penalty feature, the best result of 95.6522% accuracy on cross-validation mode has been achieved. In case of more accurate text extraction, the accuracy is even higher.

Optical Character Recognition (OCR) is a well studied subject involving variousapplication areas. OCR results in various limited problem areas are promising,however building highly accurate OCR application is still problematic in practice.This thesis discusses the problem of recognizing and confirming Bingo lottery numbersfrom a real lottery field, and a prototype for Android phone is implementedand evaluated. An OCR library Tesseract and two Artificial Neural Network (ANN)approaches are compared in an experiment and discussed. The results show thattraining a neural network for each number gives slightly higher results than Tesseract.

All dairy cows in Europe wear unique identification tags in their ears. These eartags are standardized and contains the cows identification numbers, today only used for visual identification by the farmer. The cow also needs to be identified by an automatic identification system connected to milk machines and other robotics used at the farm. Currently this is solved with a non-standardized radio transmitter which can be placed on different places on the cow and different receivers needs to be used on different farms. Other drawbacks with the currently used identification system are that it is expensive and unreliable. This thesis explores the possibility to replace this non standardized radio frequency based identification system with a standardized computer vision based system. The method proposed in this thesis uses a color threshold approach for detection, a flood fill approach followed by Hough transform and a projection method for segmentation and evaluates template matching, k-nearest neighbour and support vector machines as optical character recognition methods. The result from the thesis shows that the quality of the data used as input to the system is vital. By using good data, k-nearest neighbour, which showed the best results of the three OCR approaches, handles 98 % of the digits.

Machine detection and reading of UIC identification codes on railway rolling stock allows for automation of some processes on the railway and makes running of the railway safer and more efficient. This thesis provides insight into the problem of machine text detection and reading. It further proposes and implements a solution to the problem of reading UIC codes in line camera scanned images.

The work deals with the issue of handrwritten text recognition problem with deep neural networks. It focuses on the use of sequence to sequence method using encoder-decoder model. It also includes design of encoder-decoder model for handwritten text recognition using a transformer instead of recurrent neurons and a set of experiments that were performed on it.