DPFrag: Trainable Stroke Fragmentation Based on Dynamic Programming

Abstract

Many computer graphics applications must fragment freehand curves into sets of prespecified geometric primitives. For example, sketch recognition typically converts hand-drawn strokes into line and arc segments and then combines these primitives into meaningful symbols for recognizing drawings. However, current fragmentation methods' shortcomings make them impractical. For example, they require manual tuning, require excessive computational resources, or produce suboptimal solutions that rely on local decisions. DPFrag is an efficient, globally optimal fragmentation method that learns segmentation parameters from data and produces fragmentations by combining primitive recognizers in a dynamic-programming framework. The fragmentation is fast and doesn't require laborious and tedious parameter tuning. In experiments, it beat state-of-the-art methods on standard databases with only a handful of labeled examples.

RankFrag: A Machine Learning-Based Technique for Finding Corners in Hand-Drawn Digital Curves

“…here are DPFrag [26] and IStraw [28]. Due to the unavailability of other data, we only report the results related to the All-or-nothing metric.…”

“…The algorithms also differ for the norm they use to measure the approximation error. A recent method, called DPFrag [26] learns primitivelevel models from data, in order to adapt fragmentation to specific datasets and to user preferences and sketching style.…”

“…A set of 620 symbols was firstly introduced by Tirkaz et al [25] to measure the performance of a multi-stroke symbol recognizer. Here we use a subset of 400 symbols annotated by Tumen and Sezgin and used to evaluate a technique for finding corners [26].…”

“…It was one of the datasets used to test the homonymous technique [28] and DPFrag [26]. It contains both line and arc primitives belonging to 400 unistroke symbols, drawn by 10 different subjects.…”

“…Tumen and Sezgin [26] also emphasize the importance of the adaptation to user preferences and drawing style and to the particular domain of application. Adaptation can be achieved by using machine learning-based techniques.…”