Often a pattern recognition problem is too hard to solve with ordinary approaches involving too computationally “expensive” algorithms. Instead of this, it is sometimes better to combine the decisions coming from different simpler methods, working on the same data to produce a solution to the same problem, in order to give a global result better than that obtained by each single expert. In this paper this technique is applied to the well-known problem of handwritten digit recognition, where it is shown how the employment of a good combination applied to very simple, fast and with low single performances methods can catch up optimal reliability and performance comparable to those of “big” but slow and complex methods.
An efficient statistical approach to handwritten digit recognition
GALLO, CRESCENZIO
2009-01-01
Abstract
Often a pattern recognition problem is too hard to solve with ordinary approaches involving too computationally “expensive” algorithms. Instead of this, it is sometimes better to combine the decisions coming from different simpler methods, working on the same data to produce a solution to the same problem, in order to give a global result better than that obtained by each single expert. In this paper this technique is applied to the well-known problem of handwritten digit recognition, where it is shown how the employment of a good combination applied to very simple, fast and with low single performances methods can catch up optimal reliability and performance comparable to those of “big” but slow and complex methods.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
