Performance Analysis of Character case-sensitive and case-insensitive Classification in Handwritten Character Recognition

Abstract

In character recognition classes are letters, numbers and punctuations. Therefore, the number of classes depends on the number of characters in the language. However many classes (such as upper case “C” and lower case “c”) have very similar characteristics therefore merging such classes is also an option for a more successful recognition. In this study, we aim at evaluating the effects of abovementioned phenomenon for handwritten character recognition by performing the recognition in three different set of classes namely casesensitive (52 classes), case-insensitive (26 classes) and similarity based (38 classes) using Deep Feedforward Networks. Looking at the results, as expected caseinsensitive classification outperformed case-sensitive classification. Surprisingly, similarity based classification having a greater number of classes resulted in better accuracy rate compare to case- insensitive classification.