Reading 11: What!?! No Rubine Features?: Using Geometric-based Features to Produce Normalized Confidence Values for Sketch Recognition

Citation

Paulson, Brandon, et al. "What!?! no Rubine features?: using geometric-based features to produce normalized confidence values for sketch recognition." HCC Workshop: Sketch Tools for Diagramming. 2008.

PDF

Summary

This paper claims that geometric based features are much more effective in recognizing hand drawn sketches than gesture based features as those of Rubine's. This paper takes gestures based features i.e. the Rubine's features and geometric based features from other papers (PalioSketch) and creates a total feature set containing total 44 features. Then a subset selection of features is done to find the most effective features and it is determined that geometric based features are much more effective in making accurate decisions rather than gesture based features.

Discussion

Geometric features based sketch recognition is much more intuitive as it is more comparable to how our eyes perceive the drawn objects. When we see the final sketch we do not consider how the sketch was drawn. Though such information can further aid in classifying the sketch correctly. For example in Rubine's features a circle drawn clockwise will produce different feature values than a circle drawn counter-clockwise though the two shapes are visually similar to our eyes. This is a very important observation validated by this paper that geometric based features are much better at distinguishing sketches than gestures based features.