Skip to Main Content
Handwritten character recognition systems can be divided into three steps: preprocessing, feature extraction and classification. In the feature extraction process, the representation power of features should be increased while keeping the number of features as small as possible. We project raw character image vectors to lower dimension spaces by different linear transformations and compare their representation and discrimination power. In dimension reduction, principal component analysis (PCA), multiple discriminant analysis (MDA) and independent component analysis (ICA) are compared; the best classification performance is obtained using ICA. A multilayer perceptron, which is trained by a conjugate gradient algorithm, is used for classification. The handwritten character database studied consists of 5000 training patterns and 2500 test patterns.
Date of Conference: 28-30 April 2004