Analyzing Gestures Based on Visual Similarity

Citation

A. C. Long, J. A. Landay, L. A. Rowe, and J. Michiels, “Visual Similarity of Pen Gestures,” vol. 2, no. 1, pp. 360–367.

http://dl.acm.org/citation.cfm?id=332458

The motivation behind this paper was to find a way to determine gesture similarity. The authors conducted few experiments to find out the features that affected perceived gesture similarity. They further conducted user studies to determine the principles people use to judge gesture similarity. The data collected was used to derive metrics for predicting human perceived gesture similarity.

Trial 1
The goals of analysis was to collect metrics to devise a model for gesture similarity.
Some of the features were taken from Rubine gesture recognizer and some were added.


Trial 2 :
The goals for this trial were :

• To test the predictive power of the developed model
• systematically varying different types of features would affect perceived similarity.

In addition to this the authors wanted to measure the relative importance of features. This trial was also analyzed with the same techniques as the first similarity trial. They used Multi-Dimensional Scaling to choose the relevant features. The interesting fact is that both these trials resulted in different model. Some of them are taken from the Rubine. Some of the features which formed the model are as follows :

1. Sharpness
2. Cosine & Sine of initial angle.
3. Size and angle of bounding box.
4. Distance between first and last points.
5. Total angle and Total absolute angle
6. Curviness
7. Log Area
8. Total Angle / Total Absolute Angle 
9. Log total length
10. Log Aspect