Who Dotted That ‘i’? : Context Free User Differentiation through Pressure and Tilt Pen Data

Bibliographical Info:
Eoff, Hammond. "Who Dotted That ‘i’? : Context Free User Differentiation through Pressure and Tilt Pen Data." Graphics Interface (GI 2009), Kelowna, British Columbia, Canada, May 25-27, 2009, pp. 149--156.

URL:
http://srl.tamu.edu/srlng_media/content/objects/object-1236962325-cefe7476d664dc727f969660eac672cc/bde-GI-FinalVersion.pdf

This paper explored the concepts behind the differentiation of different writers only through the analysis of the strokes that each writes. This is decidedly different from systems where the user's identity is preserved through differentiation of the actual pen or calibration of any kind. The algorithms presented in this paper have the intention of on-the-fly user identification to make the writing experience more intuitive for the user. The paper's studies revolved around taking extensive usage metrics of different people, and running them through different classifiers to help create a threshold between which writing data belonged to which user. A number of t-tests were calculated in the first experiment to aid in the classification of users depending on the X- and Y-tilt of the user's pen. The second experiment used additional data from users and several different classifiers were used to analyze the data, including Linear, Quadratic, Naive Bayes, Decision Trees, and Neural Networks. The identification rate for two collaborating users was found to be around 97.5%.

The value presented in this paper is mostly due to the showcase in the power of data classifiers. Despite the fact that user identification when collaborating on a single surface is a problem that has been solved via a variety of different peripherals and complicated user schemes throughout the years, the solution presented in this paper is a powerful one generated exclusively through the analysis of user metrics, without the need for the user to change any of his or her writing habits. One thing to note, however, is the fact that pen tilt is one of the more important aspects of the data, which is something that not every digital pen can provide. In addition, different pressure points per stroke is also not something that pen can provide, and finger-touch interfaces were not explored. Nevertheless, this analysis is a perfect example of how running data through classifiers can be the first big step in solving complex problems such as this one.