Experiments with Adaptation Methods in On-line Recognition of Isolated Latin Characters

AbstractThe purpose of this paper is to summarize our work on adaptive on-line recog-nition methods for handwritten characters. Reports on the work have been pub-lished in various conference proceedings and book chapters. As each publicationcovers only some specific part of our work, it is hard to see the whole picture andget a good overview of the whole work. Instead of trying to explain in detail allthe techniques and experiments, we compare them with each other and give moregeneral results.By adaptation we mean that the system is able to learn new writing styles andthus improve its performance. We have had two different approaches to the adapta-tion: experiments have been carried out with both individually adaptive classifiersand adaptive committees of static classifiers.The main techniques applied in our work include the k -Nearest Neighbor andthe Local Subspace Classification rules, Dynamic Time Warpi ng and Levenshteindistances, Learning Vector Quantization, and Dynamically Expanding Context.