Handwritten Equations Solver Using Convolution Neural Network

Mathematics is fundamental to everyone’s existence. We typically use calculators embedded into our current devices to solve mathematical equations. This project aims to develop a Handwritten Equation Solver that can solve mathematical equations written by a wide range of people. A neural network approach is used to recognize symbols and alphabets. Among other things, the system can recognize alphabets, all integers, mathematical symbols, and some special symbols. When one or more variables’ coefficients are multiplied by their sum of powers, the result is an algebraic expression known as a polynomial. If x is variable, ax+b=0 then it is linear polynomial equation. We show how computers can learn cognitive skills that allow them to understand an issue based on its visual context. The problem is simplified by using an image of handwritten polynomial equations as input and producing the answer as output. The proposed workflow method automatically simplifies the Handwritten polynomial equation. Horizontal compact projection analysis and combination linked component analysis approaches are used for segmentation. To classify individual characters, a Convolution Neural Network is utilized. For the solution of the problem, a character string operation is used for each proper detection. It’s also quite good at what it does. For the solution of the equation, each proper detection, character string operation is employed. Finally, the experimental findings reveal that our proposed approach is really effective.

I. Introduction

A computer program called Handwritten Equation Recognizer [1] makes it easier to recognize the characters that make up mathematical equations. It simply entails scanning a handwritten mathematical equation into software and extracting the characters that make up the mathematical equation for further processing and calculation. The ability of a computer to recognize patterns and compute on the data obtained through pattern recognition is a remarkable accomplishment. This is a big challenge when it comes to programming a computer to understand and analyze mathematical formulas, which are a common example of pattern recognition [2]. The ability to create a neural network that can successfully "teach" a computer to recognize mathematical patterns so that it can perform further computation on the expression automatically or with minimal human input will almost certainly result in a significant reduction in the time it takes to calculate mathematical equations. In today’s scientific world, employing AI to answer a handwritten mathematical equation is a major priority. Math expressions employ a variety of sizes and two-dimensional standards. The most essential feature of this project is segmentation based on the expression’s sequence, which simplifies the arithmetic and exhibits the user’s output recognition. When a photo has been segmented, a CNN model is utilized to classify it. CNN [2] was able to remove property from the image by conducting a sequence of actions. We use a string operation to generate the expressions after the CNN model successfully recognizes each of the segmented digits and operators [3]. The square term is occasionally pushed off of the other sections of the equation’s horizontal projection curve while segmenting each quadratic from the original image. This problem is solved via compact horizontal projection [4]. Applying connected components to mathematical symbols, such as ’=’, which is a combination of two linked components, is another problem. This problem is solved using the combined linked component analysis method. The most difficult component of each handwritten character is feature extraction [5], which is tough due to the diverse shape and structure of each unique character.