Important: if you are using NVDA 2017.3 or later on Windows 10, please consider using buit-in Windows 10 OCR.

Performs optical character recognition (OCR) to extract text from an object which is inaccessible. The Tesseract OCR engine is used. To perform OCR, move to the object in question using object navigation and press NVDA+r. You can set the OCR recognition language by going to the NVDA settings panel and selecting OCR settings. The keyboard shortcut can be reassigned from NVDA input gestures dialog in the "Miscellaneous" category.