Quite tricky to get right. Parsing text from a logo is usually not so easy. So here's an alternative approach that could work: Use the first 10% of the page as the "header", no matter what it is. Store all those headers as separate image files. When you scan a new invoice/receipt, try to match it with the list of known headers. OpenCV is good for that.

Look here: http://stackoverflow.com/questions/4196453/simple-and-fast-m... http://stackoverflow.com/questions/11541154/checking-images-...

By the way, thanks for this - this has been a very useful starting point.