This research presents a novel approach to automating the generation of LaTeX code from mathematical images using advanced machine learning techniques. Given the growing complexity and volume of mathematical content, manual LaTeX creation has become increasingly inefficient. Our proposed method employs a deep neural network encoder-decoder architecture equipped with self-attention mechanisms, enabling it to accurately translate diverse mathematical notations—from handwritten to printed and scanned images—into precise LaTeX representations. To address the challenges associated with mathematical symbol identification, we meticulously curated a comprehensive dataset and implemented robust data preprocessing and cleaning techniques. Our experimental results demonstrate superior performance compared to existing methods, highlighting the effectiveness of our approach in enhancing the accuracy and efficiency of LaTeX generation. By automating this process, our methodology aims to significantly improve the accessibility and usability of mathematical notation in digital environments. This advancement has the potential to streamline workflows, reduce errors and promote broader adoption of mathematical communication tools across various fields.