Fredrik Wahlberg

A LEGO-based mobile robotic platform for evaluation of parallel control and estimation algorithms

An inexpensive robotic system intended for educational use in parallel algorithms for embedded control and signal processing is described. The hardware platform is comprised of a state-of-the-art multi-core system in a wireless network with several mobile LEGO robots that collect data from their environment. The setup covers a broad range of real-time cooperative and parallel problems arising in sensor networks, robotics, surveillance and high-performance embedded applications. As an illustration, a bearings-only tracking problem, estimating both mobile robots positions and the position of a non-cooperating target by using parallel particle filtering, is solved on the proposed platform. In order to improve the estimation accuracy and to adjust to changes in the environment and movements of the target, a controller positioning the mobile robots is utilized.

Data mining medieval documents by word spotting

This paper presents novel results for word spotting based on dynamic time warping applied to medieval manuscripts in Latin and Old Swedish. A target word is marked by a user, and the method automatically finds similar word forms in the document by matching them against the target. The method automatically identifies pages and lines. We show that our method improves accuracy compared to earlier proposals for this kind of handwriting. An advantage of the new method is that it performs matching within a text line without presupposing that the difficult problem of segmenting the text line into individual words has been solved. We evaluate our word spotting implementation on two medieval manuscripts representing two script types. We also show that it can be useful by helping a user find words in a manuscript and present graphs of word statistics as a function of page number.

Scribal Attribution using a Novel 3-D Quill-Curvature Feature Histogram

In this paper, we propose a novel pipeline for automated scribal attribution based on the Quill feature: 1) We compensate the Quill feature histogram for pen changes and page warping. 2) We add curvature as a third dimension in the feature histogram, to better separate characteristics like loops and lines. 3) We also investigate the use of several dissimilarity measures between the feature histograms. 4) We propose and evaluate semi-supervised learning for classification, to reduce the need of labeled samples. Our evaluation is performed on 1104 pages from a 15th century Swedish manuscript. It was chosen because it represents a significant part of Swedish manuscripts of said period. Our results show that only a few percent of the material need labelling for average precisions above 95%. Our novel curvature and registration extensions, together with semi-supervised learning, outperformed the current Quill feature.

Historical Manuscript Production Date Estimation Using Deep Convolutional Neural Networks

Deep learning has thus far not been used for dating of pre-modern handwritten documents. In this paper, we propose ways of using deep convolutional neural networks (CNNs) to estimate production dates for such manuscripts. In our approach, a CNN can either be used directly for estimating the production date or as a feature learning framework for other regression techniques. We explore the feature learning approach using Gaussian Processes regression and Support Vector Regression. The evaluation is performed on a unique large dataset of over 10000 medieval charters from the Swedish collection Svenskt Diplomatariums huvudkartotek (SDHK). We show that deep learning is applicable to the task of dating documents and that the performance is on average comparable to that of a human expert.

Spotting Words in Medieval Manuscripts

This article discusses the technology of handwritten text recognition (HTR) as a tool for the analysis of historical handwritten documents. We give a broad overview of this field of research, but the focus is on the use of a method called ‘word spotting’ for finding words directly and automatically in scanned images of manuscript pages. We illustrate and evaluate this method by applying it to a medieval manuscript. Word spotting uses digital image analysis to represent stretches of writing as sequences of numerical features. These are intended to capture the linguistically significant aspects of the visual shape of the writing. Two potential words can then be compared mathematically and their degree of similarity assigned a value. Our version of this method gives a false positive rate of about 30%, when the true positive rate is close to 100%, for an application where we search for very frequent short words in a 16th-Century Old Swedish cursiva recentior manuscript. Word spotting would be of use e.g. to researchers who want to explore the content of manuscripts when editions or other transcriptions are unavailable.

Feature Weight Optimization and Pruning in Historical Text Recognition

In handwritten text recognition, “sliding window” feature extraction represent the visual information contained in written text as feature vector sequences. In this paper, we explore the parameter space of feature weights in search for optimal weights and feature selection using the coordinate descent method. We report a gain of about 5% AUC performance. We use a public dataset for evaluation and also discuss the effects and limitations of “word pruning,” a technique in word spotting that is commonly used to boost performance and save computational time.

Feature space denoising improves word spotting

Some of the sliding window features commonly used in off-line handwritten text recognition are inherently noisy or sensitive to image noise. In this paper, we investigate the effects of several de-noising filters applied in the feature space and not in the image domain. The purpose is to target the intrinsic noise of these features, stemming from the complex shapes of handwritten characters. This noise is present even if the image has been captured without any kind of artefacts or noise. An evaluation, using a public database, is presented showing that the recognition of word-spotting can be improved considerably by using de-noising filters in the feature space.