FPGA Simulation of Fingertip Digit Recognition Using CNN
The paper proposes a novel framework for fingertip-writing digit recognition using Convolutional Neural Network (CNN) and its hardware implementation using Verilog. In the above-mentioned framework, an image of a handwritten digit was fed using SystemVerilog. Using SystemVerilog interfacing with C language, the image was read from a file and its corresponding RGB value was stored in a text file. Using Python, the text file was read, and the digit was recognized using CNN. CNN model was trained on MNIST dataset of 50000 images of handwritten digits with the accuracy rate of 98.4% for predicting an unseen digit. The predicted digit was displayed using SystemVerilog(Xilinx ISE). The proposed algorithm is fast in processing as it uses CNN for model training. Existing works in this field uses image classification tree to recognise digits, whose accuracy is 95.8% which is much less than the proposed algorithm in this paper.
CNN, fingertip-based digit recognition, FPGA simulation, image processing.