Recognizing English Cursive Using Generative Adversarial Networks

English cursive is present in many historical documents, and there is a need to recognize such writing and convert them to electronic documents. However, automated English cursive recognition is still considered difficult, mainly because of the segmentation difficulty presented by connected characters in a word. To resolve this issue, we propose a method to recognize whole words using GAN (Generative Adversarial Networks). This is done by using trained GAN to generate a recognizable font from cursive words, and then using conventional OCR (Optical Character Recognition) method on this font. The performance of this method is evaluated using English cursive datasets.