Reading 9: Gesture recognition features overview

Citation

Class handout (not for distribution).

Summary

The chapter covers the basic features which can be used to classify gestures or sketches. These features include the 13 features introduced by Rubine in the paper discussed in the previous reading. Apart from features introduced by Rubine 11 more features introduced by Long are mentioned in the chapter. These features are more or less derived or similar features from Rubine's features, such as Density Metric or Area of bounding box, etc. These extra features help in classifying more complex gestures but do not do much improvement in accuracy over Rubine's feature set.

Discussion

Certain features such as Area of bounding box can have dramatic effects in classification as area is a multiplicative feature and the delta's get multiplied. So a log of area is taken to map this dramatically changing feature to a less dramatic feature. Most of the features are derived features from Rubine's features. Example: One of the density metric is total stroke length divided by the length of the bounding box diagonal.