You could try Aryn DocParse, which segments your documents first before running OCR: https://www.aryn.ai/ (full disclosure: I work there).