What!?! No Rubine Features?: Using Geometric-based Features to Produce Normalized Confidence Values for Sketch Recognition

What!?! No Rubine Features?: Using Geometric-based Features to

Produce Normalized Confidence Values for Sketch Recognition

Brandon Paulson, Pankaj Raja , Pedro Davalos, Ricardo Gutierrez-Osuna, Tracy Hammond

Sketch Recognition Lab

Pattern Recognition and Intelligent Sensor Machines Lab

Spacecraft Technology Center

Texas A&M University

3112 TAMU

College Station, TX 77843 USA

{bpaulson, pankaj, p0d9861, rgutier, hammond}@cs.tamu.edu

Howdy! 

             This blog post contains important points from the above mentioned paper and a brief section of my opinion on it. 

              As I mentioned in my previous blog posts, gesture- based user interfaces are becoming more and more popular. The problem is that there are not good enough recognition algorithms to interpret a gesture correctly. Two approaches are normally used to identify a gesture: gesture- based recognition and geometry- based recognition. In this paper, the authors present a method that is a hybrid combination of those 2 methods and demonstrate the high accuracy. The algorithm is used to recognize primitive single- stroke gestures but can be easily extended to support more complex gestures by using languages like LADDER to describe complex shapes in terms of primitive ones. 

          The authors used a total of 44 features: gesture- based features from Rubine's features and geometry- based features. They use a quadratic classifier and their algorithm returns a ranked list of interpretations which can be used by a higher- level classifier to understand the context and use an appropriate interpretation. The rank is assigned based on normalized confidence values for each interpretation which is a big contribution of this paper.

         They started out with the following set of features:

Figure 1

They used a variety of feature- subsets to determine which features contributed negatively to recognition and which were very effective. The ones in the bold in the above figure were chosen to be the most important features and the set of those features proved to be most optimal in recognizing gestures.

I find the idea of using a hybrid approach very interesting and useful. It allows users to freely draw without worrying about the underlying technical details and without being constrained. 

I used the above mentioned paper as a source for this blog. Here's the link to the paper: Research Paper

Thanks for reading my blog! Have a great and blessed day!

Gig'em!!!