Building Digital Ink Recognizers Using Data Mining: Distinguishing between Text and Shapes in Hand Drawn Diagrams

The low accuracy rates of text-shape dividers for digital ink diagrams are hindering their use in real world applications. While recognition of handwriting is well advanced and there have been many recognition approaches proposed for hand drawn sketches, there has been less attention on the division of text and drawing. The choice of features and algorithms is critical to the success of the recognition, yet heuristics currently form the basis of selection. We propose the use of data mining techniques to automate the process of building text-shape recognizers. This systematic approach identifies the algorithms best suited to the specific problem and generates the trained recognizer. We have generated dividers using data mining and training with diagrams from three domains. The evaluation of our new recognizer on realistic diagrams from two different domains, against two other recognizers shows it to be more successful at dividing shapes and text with 95.2% of strokes correctly classified compared with 86.9% and 83.3% for the two others.