Interpolation/Decimation Scheme Applied to Size Normalization of Characters Images

Objective:To improve the performance of systems for handwritten character recognition. Proposed scheme: Performs size normalization (scaling), by independently interpolating/decimating the rows and columns of the original character image by any scaling factor desired, integer or non-integer. Tested interpolation/decimation algorithms: 1. Cubic Spline Interpolation, polynomial form [1], and decimation by block averaging (method proposed); 2. Cubic Spline Interpolation, polyphase network form [2], and decimation by block averaging (method proposed); 3. Yuceer normalization [3] (used for comparison). Testing environment: Recurrent Neural Networks classifier: 80 cells for input and hidden layers and 10 output cells [4]. CENPARMI Database of handwritten characters used. Results: