APPLYING HUMAN KNOWLEDGE TO IMPROVE MACHINE RECOGNITION OF CONFUSING HANDWRITTEN NUMERALS

It is believed that the decision making process of existing algorithms can be refined by gaining more human knowledge, especially when trying to recognize confusing characters. Therefore, to help improve machine performance on confusing samples, comments collected from an experiment conducted with a group of human experts specialized in unconstrained handwritten numeral recognition are analyzed. Based on this knowledge, a tool is being built which will manage the information and give results of analyses to help distinguish the different confusing styles of writing. These analyses will facilitate the design of new specialized modules aimed at differentiating numerals belonging to the same confusing pair.