Automated Labeling of Ink Stroke Data

Abstract

Labeled ink stroke data is essential to the development and evaluation of sketch recognizers. Manually labeling strokes is a tedious, time-consuming, and error prone task; and very few tools are available to facilitate this. We propose a new and intuitive method of automatic labeling for single stroke primitives. This involves building a recognizer from a partially labeled dataset. This recognizer is then used to identify and automatically label the remaining data, therefore reducing the amount of manual labeling required by researchers. An evaluation comparing manual labeling against our new auto labeling method shows that users are able to label significantly faster and produce less errors using auto labeling. Furthermore, users found auto labeling easier and more preferable.