I am a PhD student at Uppsala University working on historical handwritten text data mining and recognition. My interests include machine learning (deep learning in particular), computer vision, and natural language processing, and their intersection. My main supervisor is Anders Brun.
Tomas Wilkinson, Jonas Lindström and Anders Brun, arXiv preprint, 2017
This paper extends our previous word spotting results to the segmentation-free setting. This allows for large scale word searches in manuscript collections without requiring pre-segmented images of words.
Tomas Wilkinson and Anders Brun, ICFHR, 2016
This paper describes a model that allows you to embed word images into any general vector space in which you can perform query-by-example and query-by-string word spotting (image retrieval) and probably perform decent enough transcription of word images
Fredrik Wahlberg, Tomas Wilkinson and Anders Brun, ICFHR, 2016
We explore the use of deep ConvNets to perform manuscript production date estimation. We use a modified googlenet architecture to learn features in a supervised manner which are then used as input into Gaussian Process and Support Vector regression models.
[paper]
Tomas Wilkinson and Anders Brun, ISVC, 2015
This paper adapts a method for Object Detection (RCNN) to the new task of word segmentation. Contrary to many other word segmentation methods, this method does not segmented text lines or binary image data, while at the same time achieving results that are competetive with the state-of-the-art that do have those restrictions.
[paper]
Tomas Wilkinson and Anders Brun, ISVC, 2015
This paper outlines a procedure of how to create word clouds using image data instead of text data, including a simple way to render them to make them look reasonably good. The Image-based word cloud is a simple visualization tool that provides a quick overview of the contents of a document collection
[paper]