Hand Writing Recognition Model Enhancement Exploration Based on Skipped Connections

Handwriting recognition is a well-known problem utilizing a deep learning approach. Due to the variety of words and writing styles, a large data set is required to build the good model. In this research, we investigate the state-of-the-art handwriting model, namely Flôr model [1], with our data set, and explore the extensions to improve the accuracy. The improvement aims to use skipped connections and learn to pass through features properly. The possible skipped connection insertion approaches are studied. We also consider the model size along with the accuracy improvement. We use the modified data set originally from Research Group on Computer Vision and Artificial Intelligence INF, University of Bern [2], containing 112,7446 words. The results show that our refinement can lead to the reduction of 1.5 % in character base error and 2 % in word base error with the unseen test data set.