back to index page

Using advanced machine learning techniques to exploit spatio-temporal context for automatic stroke grouping in sketch-based interfaces.

Digital Media Research Group
University of Bremen

This work seeks to augment creative workflows in digital freehand sketching based on a contextual computing paradigm. In order to disambiguate local stroke-centric grouping decisions, Tobias Lensing and I use Echo State Networks and Support Vector Machines for time series analysis and dynamical pattern recognition. In a nutshell, this summarizes our MSc Thesis »Contextual Computing for Intelligent Systems. Using Spatio-temporal Context for Automatic Stroke Grouping in Sketch-based Interfaces«. To create a software foundation for this work, we developed »Lichtgestalt«: Lichtgestalt is an AI-assisted application for freehand sketching that runs on the Mac OS X platform, was primarily written in Objective-C/C++, has an interface to the Waikato Environment for Knowledge Analysis (Weka), and offers Python bindings through the Lichtgestalt API. We make extensive use of the Scientific Python (SciPy) package and employ a custom OS X port of the aureservoir library for the required Echo State Network functionality.

In collaboration with
Tobias Lensing.

Download Thesis (PDF, 30.1 MB)

The Lichtgestalt software is not publicly available, sorry.