OCR text going in multiple directions on one page

Is it possible to automatically recognize text going in multiple directions on the same page and duplicate the page such that there are two separately oriented copies, each with a correct OCR layer? I'm scanning thousands of scrapbook pages full of newspaper articles, and some of them are made to fit... creatively. I need every word to be searchable, and manually rotating each page can be a massive pain. (Plus, the OCR algorithm has a bad habit of orienting the page based on what it perceives as the dominant direction.)