The reliability factor based on neural networks is proposed for recognizing printed and handwritten digit characters. Five feature sets are extracted from the binary input images and assigned to the three subnetworks, each having a group of specialize...
The reliability factor based on neural networks is proposed for recognizing printed and handwritten digit characters. Five feature sets are extracted from the binary input images and assigned to the three subnetworks, each having a group of specialized feature sets respectively as their inputs. The reliability factor is applied to a modularized neural network for output voting. In winner take-all voting method, the result of subnetwork having the highest RF value is selected as output and thus the error rate can be further reduced by selecting the reliable result.