A machine learning approach to automatic stroke segmentation

Automatic understanding of sketch maps using context-aware classification

“…This starts with stroke segmentation methods (Herold & Stahovich, 2014;Huang et al, 2014) and grouping of strokes (Stahovich et al, Figure 2: Challenges for sketch map understanding (a) open street ends, (b) multi-stroke lines, (c) vague meaning: a shape that is isolated hardly recognizable, (d) vague meaning: is the middle object a lake, a lawn or a parking lot? 2014).…”

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

“…It is used to segment the shapes in diagrams of chemistry. A very recent one is ClassySeg [13], which works with generic sets of strokes. The method firstly detects candidate segment windows containing curvature maxima and their neighboring points.…”

“…There are different metrics to evaluate the accuracy of a corner finding technique. We use the following, already described in the literature [27,13]:…”

“…Thus, we just checked whether the found corner was at a reasonable distance from the marked corner. In particular, we adopted as a tolerance the fixed distance of 20 pixels already used in literature for tests on the same datasets [13].…”

A Method to Segment Strokes in Hand-drawn Sketches

“…TCVD is less susceptible to noise than the traditional curvature computation, but it is dependent on the scale which make it be confused between large radius curves and straight lines. Herold and Stahovich [9] developed ClassySeg, which employs machine learning technique to infer the corner points. A statistical classifier is used to identify which candidate segment windows contain true segment points.…”