Optical Character Recognition (OCR) technology has progressed over time from simple text recognition to modern systems that are able to process complex applications and various domains. This development has been born out of the combination of deep learning that allows for higher accuracy and flexibility for a wide range of applications. In this paper the development of a Mathematical OCR system is presented, for the recognition and extraction of mathematical equations from both handwritten and digital images. The system is based on Convolutional Neural Network (CNNs) model, making it proficient for the correct recognition of mathematical symbols and formulas. The main aim is to improve recognition accuracy of complex mathematical formulas, and to offer a step-by-step solution to each extracted formula. Trained on the CROHME dataset encompassing an ample collection of mathematical expressions, the model exhibits satisfying performance even in practice. Among the handled challenges is the symbol extraction, noise reduction and solution accuracy of noisy input images. Systemic testing demonstrates significant gains in the accuracy of OCR and computational efficiency. By integrating the pretrained model into a web application, the Math OCR system offers an interactive platform for teachers and students, revolutionizing mathematical education through automated solutions and a refined learning experience.