Reading 6: Using Entropy to Distinguish Shape Versus Text in Hand-Drawn Diagrams

Citation

Bhat, Akshay, and Tracy Hammond. "Using Entropy to Distinguish Shape Versus Text in Hand-Drawn Diagrams." IJCAI. Vol. 9. 2009.

PDF

Summary

Sketches in mechanical design or UML diagrams consist of both shapes and text. Existing systems are good at recognizing each separately but not together. The paper presents a single feature (zero order entropy) which can be used to distinguish between shapes and text with accuracy of 92.06%. The main underlying idea is that entropy (or randomness) of text strokes is much higher than shape strokes.

Discussion

The series of strokes are converted into a string of characters from (A, B, C, D, E, F, X) where A to F represent the angle a point in the stroke makes with the two neighboring points. A to F represent angles from 0 to pie. X for starting points.  This string of characters is pre-processed and entropy value is calculated. The entropy of text is visibly higher than other shapes. One question that arises is there could be shapes that are very complex, their entropy could also be higher.