Handwritten Equation Recognition with Convolutional Neural Network for Second-degree Algebraic Equations

We present in this work a method for a handwritten second-degree algebraic equations’ recognition. First, we work with a residual convolutional neural network for individual symbol recognition. Next, we merge those symbols that exceeds some overlap threshold or to have double representation. Finally, we gather all symbols that belong to the same equation using an intersect detection algorithm, by tracing multiple lines with different slops.Furthermore, we prepare an image database with more than 2000 second-degree algebraic equations handwritten by many writers. Our result shows good performance compared with other works in literature.