FPGA-based Optical Character Recognition for Handwritten Mathematical Expressions

This paper presents the hardware design of optical character recognition for handwritten mathematical expressions using field programmable gate array (FPGA). The OCR is based on feedforward neural networks. The purpose of this research is to increase the speed and reduce the energy consumption of the learning process of neural networks by performing forward and backpropagation in the hardware, instead of software. To optimize the speed and the hardware area of the design, we proposed a parallel architecture and hardware sharing design. The simulation result of our system indicated that our proposed approach achieved an accuracy of 72.5% on a 25 MHz FPGA. Based on simulation measurement, it significantly consumes a very low power, while being as fast as the baseline method in terms of wall clock time.