A Domain-Independent System for Sketch Recognition

A Domain-Independent System for Sketch Recognition

Bo Yu, Shijie Cai

State Key Laboratory of Software Novel Technology, Nanjing University, China

mikeyucn@hotmail.com

Howdy!

             In this blog post, I would like to briefly summarize the above paper and provide my view on it.

             Freehand sketching is one of the most effective way to communicate and in this paper, the authors develop a system that can recognize gestures independent of the domain and give the user freedom from using non- intuitive ways like menus and toolbars. It also allows the user to draw shapes using multiple strokes.

             The system creates a direction and a curvature graph for the input gesture and users various ways to identify the gesture and re- construct it. 

Figure 1

They use various ways to recognize a line, a circle, an arc, an ellipse and other shapes. One of the methods they use is feature area. The system calculates feature area to a line, feature area to a point and feature area to an arc as follows:

Figure 2

The vertices are detected by looking at the sharpest corners of the direction graph. The authors use various methods to eliminate false positives. Once the entire diagram has been identified, the system does post- processing which basically consists of Simple Relation Retrieval, Cleanup and Basic Object Recognition which identifies objects that are independent of the domain. The Cleanup phase (removing useless elements at the beginning and the end of the stroke) has 3 rules:

1.) If a line segment (arc) with at least one extreme point open, i.e. not connected to other elements, is very short compared with the mean length of all the elements in its original stroke, then remove this line segment (arc)

2.) Given one line segment (arc) with each extreme point connected to some other element, ifit is very short compared with its neighbor elements’ length and these two ones are neither symmetrical nor parallel, then shorten this line segment(arc)to its midpoint.

3.) If two line segments (arcs) with similar angles (centers and radii) are connected or overlapped, then merge them into a new one.

The user can select objects by doing a circle gesture around them. There are other commands that the user can provide using gestures:

Figure 3

The evaluation of the system suggests that the users liked the system. I think it's a great idea as it provides good recognition independent of the domain and can be tweaked to suit certain requirements and be made domain- specific.

I used the following sources for this blog post:

[1] Proceeding GRAPHITE '03 Proceedings of the 1st international conference on Computer graphics and interactive techniques in Australasia and South East Asia. Pages 141-146 . ACM New York, NY, USA ©2003

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

Gig'em!!!