Recoding in 5 python modules ================== - Less platform dependent implementation - Higher versality (wrt addition of new intput / output file tytes) Normalize inputs (inputs can be a pdf file, an image, a folder containing images) ---------------- - For pdf: - Identify if image already contains fonts - If the page needs to be OCRed: - Extract the image corresponding to the page and save it in a tmp folder. 3 approaches to extract images: - extract raw image from pdf and rotate it according to pdf page rotation - if not possible: identify resolution and rasterize - if not possible: use default resolution and rasterize - If not: - Save the page AS-IS in the tmp folder that should contained the final page - For image(s): - Just copy the images with standardized name into the tmp folder containing pages to be OCRed Preprocess normalized inputs (perform jobs in parallel) ---------------------------- - Orientation (if requested by user) - Correct orientation - Skew angle (if requested by user) - Correct skew angle - Cleaning (if requested by user) - Clean image Perform OCR (perform jobs in parallel) - Perform OCR and save resulting hocr file for each respective page (perfom jobs in parallel) Generate output for each page ----------------------------- - For pdf (if output file has a "pdf" extension): - Generate pdf pages from hocr files (note: pdf pages can already exist if OCR has been skipped for them) Build final output ------------------ - For pdf: - Concatenate pdf pages - Converte to pdf/1-a - Verify conformity to pdf/1-a New temp folder structure ========================= - tmp_xxxxx/ - a_raw_images (either from images or extracted from pdf file) - b_preprocessed_images (after deswing anf cleaning) - c_hocr_files - d_output_pages (one file per page (at first 1 pdf file per page. Later on other formats might be supported) - e_final_output (concatenate output pages and conversion into PDF/1-a standard, Later on support other formats)