Turn a scanned Arabic PDF into a searchable document
Identify scans, clean only the pages that need it, choose an OCR language and verify names and numbers against the source.
Try the workflow
Examples describe a practice workflow. Your file size, recognition accuracy and processing time depend on the source and your device.
Check whether OCR is necessary
Open the document in a PDF reader and try selecting a sentence. If you can select and copy meaningful text, use text extraction first. A scan often consists of page images, and a reader's search cannot find words in those pixels until a text layer is created. Mixed documents can contain both types of page.
Retain the original document and first correct the page order and rotation. The scanned-document recipe lets you organize before cleaning and recognition. Removing an irrelevant blank page can save processing time, but check that it is not a separator needed in the final document.
Clean conservatively and select the language
Grayscale, contrast and black-and-white settings can improve a faded scan, but aggressive cleanup can erase small dots or light stamps. Those dots distinguish many Arabic letters. Preview the page at a readable scale and keep the lighter original if the cleanup destroys information.
Choose Arabic, English or the available combined setting to match the pages. Start with a few representative pages. Browser OCR consumes memory and can be slow on phones; processing a smaller selected range is often easier to review than recognizing a long document all at once.
Verify the searchable output
Search the downloaded PDF for a word that appears on the source page. Copy a short sentence into a text editor and inspect its reading order. Then check personal names, dates, decimal points and reference numbers character by character. A visually good page does not prove that its invisible text layer is correct.
The review text box helps you inspect and copy recognized text. Editing that text does not rewrite the OCR layer inside the exported PDF. OCR can also make the file larger. Use the displayed output size and retain the source for checking uncertain words; this workflow is assistance with transcription, not automatic certification of a document.