The general goal of pattern recognition is to reproduce or mimic human perceptual capabilities in technical systems, making machines "see" or "hear". Generally, pattern recognition enables machines to "sense" their surroundings with a range of sensors, analyze the sensory data and react intelligently and appropriately to certain events occurring in these surroundings. Relevant events are associated with reappearing patterns in the sensory data streams. Thus, the task is to find, model (or "learn") and classify those patterns, distinguishing relevant from irrelevant events.
Research in the Pattern Recognition Group aims both at advancing the principled pattern recognition methods behind Intelligent Systems and at developing application-oriented solutions for real-world problems. The term Intelligent Systems comprises a wide range of artifacts and devices augmented with advanced computational capabilities. Good examples are robotic assistants, smart homes, or decision support systems. All these have in common that they react intelligently when interacting with humans and draw "intelligent" inferences in automated decision processes. In order to realize such seamingly intelligent behavior advanced techniques of pattern recognition and machine learning are developed and applied.
Currently, the main research topics addressed come from the fields of computer vision, acoustic signal processing, and document image analysis. In these areas techniques for the natural and robust interaction between technical systems - like, for example, smart spaces - and human users are developed. The solutions are primarily based on the application of advanced methods from the field of statistical pattern recognition. These all share the important property of being able to automatically learn computational models from examples, which is also a fundamental capability of human perceptual systems.
The group closely collaborates with several research groups from academia and a number of industrial partners.
For motivated students, we always offer a wide variety of topics for student work and Bachelor/Master/Diploma Theses in the fields of pattern recognition methods and applications.
Prof. Gernot A. Fink von der Fakultät für Informatik forscht zu Handschrifterkennung. Als General Chair hat er gemeinsam mit einem internationalen Team die 17. International Conference on Frontiers in Handwriting Recognition (ICFHR) 2020 organisiert, die Anfang September in Dortmund und damit erstmals in Deutschland hätte stattfinden sollen. Aufgrund der Corona-Pandemie wurde sie jedoch online durchgeführt. Im Interview berichtet Prof. Fink, wie Handschrifterkennung und Informatik zusammenhängen und warum die Handschrift auch im digitalen Zeitalter nicht aussterben wird.
The 14th International Workshop on Document Analysis Systems (DAS 2020) was held July 27-29, 2020. Originally planned to take place in Wuhan, China, the workshop was held in a virtual format. The program focuses on system-level issues and approaches in document analysis and recognition and it comprises invited speakers, oral and poster presentations.
This year the Nakano Award honouring the best paper was awarded to
Fabian Wolf and Gernot A. Fink
from the Pattern Recognition Group, Department of Computer Science, TU Dortmund, Germany for their paper
Annotation-free Learning of Deep Representations for Word Spotting using Synthetic Data and Self Labeling.
The authors continued the successful works of the pattern recognition group on the topic of word spotting and presented a novel training scheme. A big limitation on the application of word spotting systems based on machine learning is their high demand for manually labeled data. In their work, the authors were able to show that it is feasible to to train a neural network solely on a synthetic dataset and still achieve high retrieval performances.
The Pattern Recognition Group is one of the founding partners of Europe's first "Time Machine". The European Commission has chosen Time Machine as one of the six proposals retained for preparing large scale research initiatives to be strategically developed in the next decade. €1 million in funding has been granted for preparing the detailed roadmaps of this initiative that aims at extracting and utilising the Big Data of the past. Time Machine foresees to design and implement advanced new digitisation and Artificial Intelligence (AI) technologies to mine Europe’s vast cultural heritage, providing fair and free access to information that will support future scientific and technological developments in Europe.
The Time Machine will create advanced AI technologies to make sense of vast amounts of information from complex historical data sets. This will enable the transformation of fragmented data – with content ranging from medieval manuscripts and historical objects to smartphone and satellite images – into useable knowledge for industry. In essence, a large-scale computing and digitisation infrastructure will map Europe’s entire social, cultural and geographical evolution. Considering the unprecedented scale and complexity of the data, The Time Machine’s AI even has the potential to create a strong competitive advantage for Europe in the global AI race.
Cultural Heritage is one of our most precious assets, and the Time Machine’s ten-year research and innovation program will strive to show that rather than being a cost, cultural heritage investment will actually be an important economic driver across industries. This constant source of new knowledge will be an economic motor, giving rise to new professions, services and products in areas such as education, creative industries, policy making, smart tourism, smart cities and environmental modelling. For example, services for comparing territorial configurations across space and time will become an essential tool in developing modern land use policy or city planning. Likewise, the tourism industry will be transformed by professionals capable of creating and managing newly possible experiences at the intersection of the digital and physical world. These industries will have a pan-European platform for knowledge exchange which will add a new dimension to their strategic planning and innovation capabilities.
The 16th International Conference on Frontiers in Handwriting Recognition (ICFHR 2018) has been held August 5-8, 2018 in Niagara Falls (USA). ICFHR is the flagship conference for handwriting recognition research and applications. The conference brings together experts from academia and industry to share their experiences and advance research in all aspects of handwriting recognition technologies. During the conference, outstanding contributions are recognized by awards for the best paper, the best student paper, and the best poster sponsored by IAPR, the International Association for Pattern Recognition. This year, the IAPR Best Student Paper Award goes to
Eugen Rusakov, Leonard Rothacker, Hyunho Mo, and Gernot A. Fink
from the Pattern Recognition Group, Department of Computer Science, TU Dortmund, Germany for their paper
A Probabilistic Retrieval Model for Word Spotting based on Direct Attribute Prediction.
In this paper the authors present an extension of the successful deep learning method for word spotting developed in the same research group – the so-called PHOCNet – that significantly improves retrieval performance on challenging word spotting tasks by casting the matching procedure between queries and target words into a probabilistic framework.
Eugen Rusakov at the award ceremony.
In the context of the project "Computer-assisted Cuneiform Analysis" (CuKa), the German Research Foundation (DFG) funds a collaboration between the Academy of Sciences and Literature, Mainz, and the Department of Computer Science at TU Dortmund (Computer Graphics and Pattern Recognition groups). The goal of the project is the development of an analysis system for cuneiform tablets, that will provide functionality for automatically searching for symbols and words to humanities scholars.
Ubiquitous systems are becoming an integral part of our everyday lives. Functionality and user experience of these systems often depends on accurate, sensor-based activity recognition and interaction. The objective of the International Workshop on Sensor-based Activity Recognition and Interaction (iWOAR) is to discuss challenges and possible solutions in this field.
Every year the workshop awards the best paper. This year 2017's iWOAR best paper award was earned by René Grzeszick, Jan Marius Lenk, Fernando Moya Rueda and Gernot A. Fink from the Pattern Recognition in Embedded Systems Group in collaboration with their colleagues Sascha Feldhorst and Michael ten Hompel from the Fraunhofer IML/the chair of Materials Handling and Warehousing, Faculty of Mechanical Engineering, TU Dortmund. They were awarded for their work entitled: Deep Neural Network based Human Activity Recognition for the Order Picking Process.
The paper discusses a novel approach to recognizing human activities in industrial settings using a deep neural network. The approach has evaluated using data from multiple warehouses.
By tradition, the location of the International Conference on Frontiers in Handwriting Recognition (ICFHR) is decided by the conference participants four years in advance of the respective event. The basis for the voting are bids for hosting submitted by tentative organization teams. For ICFHR 2020 two teams presented hosting proposals at this year's ICFHR 2016 in Shenzhen, China, namely for Vienna and Dortmund as possible locations of the venue. In a crucial vote, the Dortmund team headed by Professor Gernot A. Fink prevailed against the competitors from Vienna. Therefore, ICFHR will come to Germany for the first time in its history and will be hosted by TU Dortmund University in September 2020.
This years International Conference on Frontiers in Handwriting Recognition (ICFHR) proved to be full of awards for the Pattern Recognition in Embedded Systems Group. First, Sebastian Sudholt, Leonard Rothacker and Gernot A. Fink won the Query-by-String track in the Handwritten Key Word Spotting Competition. Furthermore, Sebastian Sudholt and Gernot A. Fink earned best paper honors for their work entitled PHOCNet: A Deep Convolutional Neural Network for Word Spotting in Handwritten Documents.
In this paper a Deep Learning approach for word spotting was presented which achieves state-of-the-start results across a number of challenging word spotting tasks thereby outperforming previous methods by a large margin. The same method was also used in the groups contribution to the word spotting competition.
Sebastian Sudholt and Gernot A. Fink presenting the awards
The Pattern Recognition in Embedded Systems Group has won the training-free track of the 2015 Key Word Spotting Competition held during the 13th International Conference on Document Analysis and Recognition. The method presented by Leonard Rothacker, Sebastian Sudholt and Gernot A. Fink was able to largely outperform their opponent's approaches in a segmentation-free and a segmentation-based scenario.
Key Word Spotting is the problem of retrieving word images from a document image collection relevant to a specific query. In the training-free track of the 2015 Key Word Spotting Competition word images were used as queries which is commonly referred to as Query-by-Example. The track was further subdivided into two assignments: for the first assignment a perfect segmentation for each word image in the document image collection was given while for the second task entire document pages where presented with no segmentation whatsoever. In both assignments, the Pattern Recognition in Embedded Systems Group achieved the highest score and consecutively claimed first place in the training-free track. A complete evaluation of the competition can be found here while the method presented is described in detail here.
The award winning team from left to right: Leonard Rothacker, Sebastian Sudholt, Gernot A. Fink
Winner of the p5 award 2013 is PG 568: "TabScript - Handwriting recognition on Android based tablets". The award is sponsored by the Alumni Computer Science Dortmund and given to student projects with a high practical relevance. In a very close voting between five interesting projects the TabScript group was able to convince the audience of the DAT2013 (Day of the Alumni Computer Science Dortmund) by the brave presentation of a live demo.
Over the last year the students developed a notepad application for Android tablets that supports handwritten input instead of the softkeyboard. Using Hidden-Markov-Models the handwritten text is recognized and translated into machine readable text. The ESMERALDA toolkit is used for implementing the Hidden-Markov-Model. A character model is trained on data from Unipen database that is publicly available for research purposes. In addition, different dictionaries with about 2500 words are used in order to recognize complete handwritten words. Dictionaries containing the most common words in German and English were included in the application. In case studies the trained models showed a word recognition rate of around 79%.
Students: Sulejman Begovic, Rebecca Doherty, Shinazi Faruki, Daria Filatova, Nina Hesse, Dennis Kesper, Julian Kürby, Niclas Raabe, Johann Straßburg, Christian Wieprecht
p5 projectgroup award
Ifran Ahmad, M. Sc., has been awarded with the "IAPR Best Poster Award" for the joint publication "KHATT: Arabic Offline Handwritten Text Database". In collaboration between King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia, the Pattern Recognition in Embedded Systems group and TU Braunschweig, Germany, a large Arabic handwriting database was created.
Best Poster Award Ceremony
(from left to right: Masaki Nakagawa, Bidyut Baran Chaudhuri, Volker Märgner, Irfan Ahmad,
Gernot A. Fink, Sebastiano Impedovo)
On April 11, 2012 the movie shooting for the new Tatort Dortmund started in the TEC Center Colani in Lünen. It was the first shooting outside the studio. The Tatort is a famous German TV detective-series that first aired in 1970. The Tatort Dortmund theme, its filming locations and the references to next generation technologies are reflecting the structural change in the Ruhr-Area. Being formerly known for its mining and steel facilities, it has become important in the high-tech sector.
The service robot of the Pattern Recognition Group (Department of Computer Science 12) was participating as a supporting actor. It's task was to welcome the detectives in a fictional high-tech-company.
Tatort shooting at TEC Center Colani
A short TV report about the shooting can be found in the WDR Mediathek (German):
http://www.wdr.de/mediathek/html/regional/2012/04/11/lokalzeit-dortmund-tatort.xml
Akmal Junaidi, M. Sc., has been awarded with the "Best MOCR Student Paper Award" on the "Joint Workshop on Multilingual OCR and Analytics for Noisy Unstructured Text Data" in Bejing, China. The workshop was held in conjunction with the renowned "International Conference on Document Analysis and Recognition" (ICDAR). Mr. Junaidi is conducting his PhD research at LS XII of the Faculty of Computer Science in the field of automatic analysis and recognition of handwritten documents in Indonesian Lampung script, and is supported by a grant from the Indonesian government. The award-winning work is entitled "Lampung - a New Handwritten Character Benchmark: Database, Labeling and Recognition".
Prize winner Akmal Junaidi (middle) with his co-authors Szilárd Vajda (left) and Gernot A. Fink.
