Improved handwritten character recognition thanks to a new geometric distortion method

A new character distortion method for off-line hand-written character recognition is presented. By allowing to artificially create new characters images from real ones, this method can be applied to increase the diversity of the database that is used for training a classifier, which can then results in a significant improvement of its generalisation ability. The principle of the proposed method is to apply linear and geometrical distortions in combination, to a bidimensional sampling grid, which is then used to resample the character image. The proposed method only depends on a few number of parameters, which leads to these ones being very easily set out, so as to ensure that enough new useful information will be provided to the classifier, as well as to avoid creation of over-noisy images. The tests that were carried out on hand-written digits extracted from the NIST3 database (1) have shown that this method allows to reduce significantly the misclassification error rate: by training so a Multilayer Perceptron as a classifier, the recognition rate obtained on an independent test set has been increased from 97.0% to 98.1%.