Pengenalan Simbol Matematika dengan Metode Convolutional Neural Network (CNN)

Abstract

Handwriting recognition of mathematical symbol has problem in the field of pattern recognition and that make more difficult to detect than only handwriting. The complex structure of writing and diverse mathematical symbol makes handwriting of mathematical expressions very difficult to process the segmentation, symbol recognition, and structure analysis. The process is used to convert handwritten mathematical expressions into digital text format such as LaTeX or MathML. In this case, the process of symbol recognition becomes the focus of research, by comparing models contained in the Convolutional Neural Network method and find the greatest accuracy. Convolutional Neural Network (CNN) is one of the methods that can be used to recognize handwriting very accurately. In this research CNN will be used to recognize symbols in mathematical expressions and will be tested with CNN models (ResNet34 and DenseNet121). This research also explains how the deep learning approaches, such as CNN that can distinguish an object in an image, and after the two models have been tested we learn that both models have different performance and architecture, the DenseNet121 model becomes a better model when compared to ResNet34 in accuracy.