Study of the Number Recognition Algorithms Efficiency After a Reduction of the Characteristic Space Using Typical Testors

The present work studies an application of typical testors to reduce the space of characteristics used as inputs in statistical models to recognize a manuscript digit from its corresponding binary image. The fundamentals of typical testors are described, and the use of logistic and multilogistic regression models in the predictions. Images from the public database of the Semeion Research Center of Sciences of Communication are used. Adequate prediction results were achieved in a reduced space of characteristics.