Orthogonal LDA in PCA Transformed Subspace

The paper addresses the effectiveness of orthogonal linear discriminant analysis (OLDA) in a principal component analysis (PCA) transformed subspace. The performance of the technique is studied for writer independent recognition of online handwritten Kannada numerals. Experiments show that the performance of LDA and OLDA are better in a PCA transformed subspace compared to that of the original feature space. In addition, the recognition accuracies of the system with OLDA are marginally better than that of LDA in both the original feature space and the PCA transformed subspace. An average recognition accuracy of 96.9% is achieved on a database collected from 69 writers. To our knowledge, this is the first ever reported work on recognition of online handwritten Kannada numerals.