Seminars at CBA

Closing in on my PhD half-time point, I will give a brief summary of my work during 2019 - the development of a tool to cluster and visualise word and character image embeddings and how this can be extended to other types of images and metadata - followed by a report of my current research project. The latter is focusing on removing strike-through marks from handwritten text to ease further processing.

I will start by defining Metric, Algorithm and Error for distance transforms, as a recent guest confused these terms quite a bit. Then I will take the example of Skeletons from distances. Skeletons are computed from shapes or backgrounds to reduce their dimensions and thus make many operations, such as shape manipulation, simplification, and segmentation easier and cleaner. In continuous geometry the three different common definitions of skeletons are equal, but in digital geometry they usually give different skeletons. This is independent on the metric used. So, remember that even not to choose is a choice and thus choose consciously.