Scale-space based feature point detection for digital ink

Abstract

Feature point detection is generally the first step in model-based approaches to sketch recognition. Feature point detection in free-hand strokes is a hard problem because the input has noise from digitization, from natural hand tremor, and from lack of perfect motor control during drawing. Existing feature point detection methods for free-hand strokes require hand-tuned thresholds for filtering out the false positives. In this paper, we present a threshold-free feature point detection method using ideas from the scale-space theory.