Analysis of Different Subspace Mixture Models in Handwriting Recognition

In this paper we explore, analyze and propose the idea of subspace mixture models such as Principal Component Analysis (PCA), Fisher´s Linear Discriminant Analysis (FLD) and Laplacian in handwriting recognition. Statistically, Gaussian Mixture Models (GMMs) are among the most suppurate methods for clustering (though they are also used intensively for density estimation). By modeling each class into a mixture of several components and by performing the classification in the compact and decorrelated feature space it may result in better performance. To do this, each character class is partitioned into several clusters and each cluster density is estimated by a Gaussian distribution function in the PCA, FLD and Laplacian transformed space. The analysis of different mixture models are experimented out on handwritten Kannada characters.