OCR Word Document Files That Are Pictures of Pages

To OCR a Word document whose pages are pictures of a scan, get the pictures out of the DOCX first, by saving the file as a PDF or extracting the images. Run OCR on that PDF or those images, export real text, and build a new Word file from it. Keep the file you were sent unchanged.

Why would a Word document need OCR?

A Word document needs OCR when its pages are pictures. Someone scanned or photographed paper, inserted each page image into a blank document and saved it. The file has a .docx name and opens in Word, but it contains no typed characters, only images that show characters.

A file can reach you this way for ordinary reasons: a client asked for "a Word copy" had only paper and a phone, or a colleague dragged scans into a document to keep them together. Whatever the reason, the file cannot be edited, searched or compared until its pictures are read.

The mechanics of why pasting does not convert anything are explained in the post on JPG conversion, under why a picture pasted into a Word file is still not text. What matters here is what to do with the file once you have one.

One point about tools. Check which file types your OCR tool takes as input. Schema Law, for example, accepts uploads as PDF, PNG or JPEG, so a DOCX has to be turned into one of those before it can be read. That preliminary step is what this post is about.

How do you tell pasted page pictures from typed text in a DOCX?

Look at the word count in Word's status bar. A document of twelve full pages that reports zero words, or only a handful, is made of pictures. Confirm it by pressing Ctrl+A and changing the font: typed text changes appearance, and pictures of text do not move.

Three further checks, each a few seconds:

• Arrow through a page. Click at the top of a page and hold the down arrow. In typed text the cursor walks line by line. On a picture page it jumps from one page to the next, because each page is a single object.
• Look at the file size. A dozen pages of typed text make a small file. A dozen page pictures make one many times larger. Size alone proves nothing, but a 30 MB engagement letter is worth a second look.
• Try to place the cursor mid-sentence. If you cannot put a blinking cursor between two words, there are no words there.

Do the check on every page, not the first alone. A typed cover page in front of pictured pages is exactly the file that fools a quick glance.

Getting the page pictures out of a Word file so OCR can read them

There are two ways out: save the Word file as a PDF, which turns each pictured page into a PDF page, or extract the image files from the DOCX. Saving as PDF is quicker and keeps the pages in order. Extracting gives you the images as they are stored in the file.

Route one: save as PDF

Work on a copy of the file. Microsoft's instructions for saving or converting to PDF in Office desktop apps give the steps for Word: select File, then Save a Copy, choose where to save, pick PDF in the file type list and select Save. The result is a PDF whose pages hold the same pictures in the same order.

Before you upload it, page through the PDF. Check that each picture fills its page and that none was cut off by Word's margins or pushed on to a following page. If a picture was pasted small, with wide white space around it, enlarge it in the copy to the width of the page and save the PDF again, so the letters occupy more of the page.

Route two: extract the images

A DOCX is a zip package. Microsoft's page on Open XML formats and file name extensions says the format "uses zip compression technology to store documents" and that "All you need to open and edit an Office file is a ZIP utility and an XML editor."

1. Make a copy of the DOCX and change the copy's extension to .zip.
2. Open the zip and look inside its word folder for a media folder.
3. Copy the image files out to a working folder.

Do not assume the file names give the page order. Open each image, note which page it shows, and rename the files with page numbers. With more than a few, join them into one PDF before OCR; the post on turning scanned images into one PDF covers ordering and combining.

Image quality either way

Pictures inside a Word file may be softer than the scans that went in. Microsoft's article on reducing the file size of a picture in Office states that "The default picture resolution in Microsoft 365 is 220 ppi." If the print is small and the recognized text comes back poor, ask the sender for the original scans or the paper.

The Okafor file is a set of supply terms. Proof the warranty, any excuse of performance and the delivery dates against the pictures. If the pages cite the Uniform Commercial Code, keep that citation with the clause it sits in, because a page-picture file often splits the sentence.

Putting recognized text back into the Word file in place of the pictures

Do not put the text back into the file you received. Build a new Word document from the OCR output, proof it against the pictures, and keep the picture file as the record of what was sent. Replacing pictures inside the original destroys the only copy that shows what the pages looked like.

The sequence:

1. Run OCR and export Word. Upload the PDF or images and export a DOCX. In Schema Law the Word export is described on the formats list as editable text with its headings, lists and tables.
2. Name the two files so they cannot be confused. For a fictional matter: Okafor supply terms (as received, page pictures).docx and Okafor supply terms (OCR conversion, unchecked).docx.
3. Proof the conversion against the pictures. Open the two side by side. Read names, dates, amounts and clause numbers against the images, and correct recognition errors in the new file only.
4. Rename when checked. Replace "unchecked" with the checker's initials and the date.
5. Tell the sender what you did, if it matters. If the document is going back with changes, say that the editable version was produced by OCR from their pictures, so they can check it against their own original.

If a single document must hold both, put the text first and the page pictures after it as an appendix headed "Pages as received". The pictures then stay available for comparison without sitting between the paragraphs.

Word files that mix typed paragraphs with pasted scans

In a mixed file, leave the typed paragraphs alone and convert only the pictures. The typed text is exact already. Save each pasted scan as an image, run OCR on those images, and type or paste the recognized words into a working copy next to the picture they came from.

Take a fictional witness statement prepared for Lena Brandt. Its nine typed paragraphs are real text. Between paragraphs 4 and 5 sits a pasted picture of a delivery note, and after paragraph 9 a pasted picture of a two-page letter. Saving the whole file as a PDF and recognizing everything would turn the nine good paragraphs into recognized text for no benefit.

Instead:

1. List the pictures. Scroll through and note each one and the page it sits on. Three images in this file.
2. Save each picture out. Right-click the picture and choose the option to save it as a picture, as PNG. Name it for its position: Brandt statement - image after para 4 (delivery note).png.
3. Run OCR on the images. Export plain text.
4. Check each text against its picture.
5. Place the text in a working copy. Under each picture, add the recognized text in a clearly labelled block, such as "[Text of the delivery note pictured here, from OCR, checked 7 Oct 2026]". Leave the picture where it is.

Whether a transcription belongs inside a statement or exhibit at all, as opposed to your own working copy, depends on the rules and practice that apply to that document. Ask the lawyer responsible before anything is added to a version that will be signed or filed.

The same split, exact text on one part of the page and words trapped in an image on another, turns up in PDFs too. The post on how to convert PDF images to text covers finding and filling those gaps.

For a first test, Schema Law's free plan takes one document a day of up to three pages, and the pricing page shows the limits for longer files.