AI Equation Solver through OCR

During 2017, as a proof of concept of AI, we built an optical equation solver. Using a camera and computer vision (OpenCV), we built an app which detected written letters and signs real-time. A multilayered perceptron neural network would then classify an image chunk as a character using optical character recognition (OCR), which could then be used in our equation solver. With this solution, a user can write close to any expression or equation containing known operators and let the application compute an answer.

In short, the process is divided into three parts:

1. Preprocessing - Each image in a read dataset contains a character which is converted into a string of bits by thresholding the image. The image is then cropped and rescaled before storing the result into a text file and which is used in the following step

2. Training - The processed dataset is used as the input for the training of the MLP-neural network. The result is exported and used in the application.

3. Application - The application uses a camera to register characters. The neural network guesses the classification of the characters in each frame. If the registered characters define a valid mathematical expression, the application calculates the result of the expression and dispose it to the user.

Todays AI application and usage of LLM’s provide a much broader and capable framework, however the basis of what makes the algorithm tick remains the same. We understand the underlying concepts, what it needs to learn proficiently and how it can be utilized to maximum effect.