chungkwong/MathOCR — reverse-engineered prompt

Build me a Java desktop app that can read printed scientific document images and turn them into editable results, especially math formulas. I want to open common image files like PNG, JPEG, GIF, BMP, and PNM, preview the page, run cleanup steps like grayscale, thresholding, noise removal, border cleanup, and skew correction, then have the app detect the layout, reading order, text blocks, tables, images, normal text, and equations.

The important part is recognizing mathematical symbols and formula structure without requiring another OCR program. It’s fine if optional external OCR engines can be connected later. Let users export the result as plain text, LaTeX, or HTML with MathML. Please include a simple GUI and also make it possible to run from a built jar.

This can be an early research quality prototype, so accuracy doesn’t have to be perfect, but keep the code organized so recognition models, segmentation methods, and output formats can be improved later. Use Java and Maven, and look up current docs online if needed.