Memory conscious sketched symbol recognition

Abstract

Automatic sketch recognition is used to enhance human-computer interaction by allowing a natural/free form of interaction. It is a challenging problem due to the variability in hand drawings, the variation in the order of strokes, and the similarity of symbol classes. Since sketch recognition requires real time processing, the speed of the classifier is important. Another important issue is how to deal with very large data sets and/or large number of classes, as these also effect training and testing speed, making certain approaches infeasible. In order to deal with these issues, we present a memory conscious sketch recognition system that processes the data to retain only a few templates per class as prototypes; and furthermore, the query and prototypes are subsampled without loosing important information. The system also uses a cascaded combination of classifiers, to improve speed, as well as recognition accuracy. Results obtained using the public COAD and NicIcon databases are comparable to previous results obtained for these databases.