Reading 5: An image-based, trainable symbol recognizer for hand-drawn sketches

Citation

Kara, Levent Burak, and Thomas F. Stahovich. "An image-based, trainable symbol recognizer for hand-drawn sketches." Computers & Graphics 29.4 (2005): 501-517.

PDF

Summary

Paper presents a new way to recognize hand-drawn figures in a sketch. The paper uses image-based approach. Achieve rotational in-variance (using polar coordinates), which is difficult in template matching systems. Current systems achieve rotation in-variance but at high cost which is not suited for interactive use. The system requires single prototype example to learn an new symbol. 

Discussion

First the definitions which are markedly dissimilar using polar coordinates are pruned. The remaining are passed through 4 classifiers, the results of which are pooled and final decision is made. The 4 classifiers use Hausdorff Distance, Modified Hausdorff Distance, Tanimoto Similarity Coefficient and Yule Coefficient. The output of the 4 classifiers is different, they are converted into a similar unit of measurement, normalized and then combined.