CE-06.15

Title: 

Incorporating Global Sketch Features into Digital Ink Clustering

Author(s): 

Chunyong Fei

Oral Defence Date: 

12/22/2006

Location: 

TH 906

Committee: 

Prof. Dragutin Petkovi, Prof. Ilmi Yoon, Prof. Edward Lank

Abstract: 

Free-form sketching is a nature and efficient method during informal tasks such as note taking and early stage design drawing. Most previous applications that support free-form sketching rely on prior constrains from a particular domain or, in the absence of a domain, simply apply character recognition. SFSU student Yi Li has developed an ink clustering algorithm that operates on generic sketches regardless pre-defined domains, and performs sketch visual structure detection based on the Spatial Proximity feature from principles of Perceptual Organization. However, the limited deployment of features from Perceptual Organization largely affected the accuracy and usability of her clustering algorithm. The main purpose of our project is to extend Yi Li’s algorithm to utilize more features from Gestalt Laws and group the unorganized digital ink into more visually meaningful structure. Our extended features include Temporal Ordering, Continuity/Alignment and Containment. Furthermore, some techniques and calculations in Yi Li’s program needs to be refined and re-implemented. Because of the increasing popularity of Table PC and the highly flexible and easy of use .NET developing environment, our whole project is implemented with C# and Microsoft Tablet SDK. The framework for grouping of free-from ink expressions allows a user to select his/her targeted clustering feature through a feature selecting form, and performs ink clustering and display through the interaction with Tablet PC SDK. Domain independent free-form sketching experiments demonstrate that the system can achieve a satisfactory segmentation/clustering accuracy.

Keywords: 

Digital Ink, Ink Structuralizing, Pen-based computing, Text Line Extraction, Spatial Proximity, Temporal Ordering, Smooth Continuity, Containment.

Copyright: 

Chunyong Fei