Handwritten numeral recognition by mirror image learning for autoassociation neural networks

摘要

This paper studies on the mirror image learning algorithm for the autoassociative neural networks and evaluates the performance by handwritten numeral recognition test. Each of the autoassociative networks is first trained independently for each class using the feature vector of the class. Then the mirror image learning algorithm is applied to enlarge the learning sample of each class by mirror image patterns of the confusing classes to achieve higher recognition accuracy. Recognition accuracy of the antoassociative neural network classifier was improved by the mirror image learning from 98.76 to 99.23 in the recognition test for handwritten numeral database IPTP CD-ROM1 1.