You can indeed do OCR at any stage. The only potential issue I see is that JPEG compression (as used by PDF) might interfere with some OCR algorithm and make it

You can indeed do OCR at any stage. The only potential issue I see is that JPEG compression (as used by PDF) might interfere with some OCR algorithm and make it less reliable, so OCRing directly on the raw output of the scanner might yield a somewhat better recognition. Then again, I know nothing of the protocols used by scanners over USB; it might well be that scanners send their result already compressed, or that the resolution is so high compared to the letter size that this doesn't matter.