Paper Review #4: Specifying Gestures by Example

1. Paper Bibliography

• Title: Specifying Gestures by Example 
• Authors: Dean Rubine
• Publication: Vol. 25. No. 4. ACM, 1991.

2. Summary

• This paper tested GRANDMA program which help to implement hand-gesture application. They used GDP as hand-gesture application. With simple gesture(click and drag) and single stroke, they defined the function of each gesture and design them with GRANDMA. The main steps are collecting data, classifying them and manipulation. For evaluation, they tested varying the number of class and training examples. As a result, in spite of its simplicity, it performed well. This paper suggested that this simple step will be a big stepping stone for future work.

3. My opinion

    1) Method

• They suggested classification algorithm and training algorithm which are short and simple. Basically, since the data are two dimensional, single-stroke gestures, algorithm might be relatively simple. They extracted features from the data and classified the features as a certain class. Then, they ran the training data and rejected some ambiguous data. In these days, this procedure might be well-known, but considered the time the paper published, I think this established the standard for classification. 

    2) Idea

• I think this idea is pretty simple except all the equations. However, I am impressed the results. In Figure 9, all sets got over 98% of accuracy.

    3) New research idea

• When I read only the title of the paper, I thought the gestures includes various thing such as hand motions, body movement. Then, I realized that this has been published a quite long time ago. I did not come up with new idea but I can tell that this paper is a milestone for sketch recognition techniques.