Scanned Image to Text: PNG, JPG, TIFF and Screenshots

To turn a scanned image into text, give the image file to an OCR tool and export what it reads as plain text. PNG and JPG files usually go in as they are. TIFF, HEIC and other types may need saving in an accepted format first. Results depend on the pixels in each letter more than on the file type.

Image files reach a matter from every direction: a TIFF from a county recorder's system, a PNG screenshot of a client portal, a JPG a witness took of a notice on a door. Each holds words you can see and cannot select. The job is the same for all of them, and the handling differs in small ways that decide whether the text comes out right.

Which image file types can OCR read as text?

OCR can read text from any image type once the software has opened it as a grid of pixels, so the real question is which types a given tool accepts. JPEG and PNG are taken almost everywhere. TIFF is common in scanning software and less so online. HEIC, WebP, BMP and GIF vary from tool to tool.

Schema Law accepts three upload types: PDF, PNG and JPEG, each up to 20 MB. Everything else has to be saved as one of those first.

Two columns in that table deserve a closer look.

Lossy or not. A lossy format discards detail to make the file smaller, and it does so again each time the file is saved. A lossless one keeps every pixel. For text, lossless is safer, though a lightly compressed JPEG reads perfectly well. Converting a lossy file to a lossless type afterwards does not bring the detail back.

Pages per file. Most image types hold one picture. TIFF and PDF can hold many, which is why a forty-page deed packet may arrive as a single .tif. Open it in a viewer that shows a page count before you assume it is one page.

The format tells you how the pixels are stored. It says nothing about whether they show legible print, so a PNG of a blurred page reads no better than a JPG of it. If what you want from a picture is a formatted, editable document instead of bare text, that is a different export, covered in the post on using OCR to take a JPG to Word.

TIFF, HEIC and other image files that need converting first

When a tool will not take your image type, open the file in a viewer or image editor and save a copy in a type it does take. Convert once, from the original, at full size and without added compression. Then keep the original, because the converted copy is a derivative and the file you received is the source.

Changing the file's extension does not convert it. Schema Law, for one, decides what a file is by examining its contents, so a HEIC photo given a .jpg name is still refused. The conversion has to be a real save or export.

TIFF

A single-page TIFF converts cleanly to PNG, which keeps every pixel. A multi-page TIFF should become a PDF, so the pages stay together in order. Most image viewers and PDF programs can open a TIFF and print or export it to PDF.

Check three things afterwards:

• Page count. A viewer that only shows the first page of a multi-page TIFF will happily export just that page. Compare the count before and after.
• Dimensions. Look at the pixel size of a page before and after. An export that quietly halves the resolution has cost you half the detail in each letter.
• Black and white pages. Many TIFFs from fax and records systems are pure black and white with no greys. They are small and usually read well. Make sure the conversion did not turn them into heavily compressed colour JPEGs inside the PDF.

HEIC

An iPhone or iPad set to the High Efficiency camera format saves photos as HEIC. Apple's support page on HEIF and HEVC media on Apple devices explains that you can switch the camera to JPEG under Settings, Camera, Formats by choosing Most Compatible, and that media shared by AirDrop, Messages or email might automatically be sent in a more compatible format such as JPEG. So a client's photo may already be a JPEG by the time it reaches you, or it may not. Look at the extension, and if it is .heic, export a JPEG at full size from a photo program.

Tell clients who will be sending many pictures about the Most Compatible setting. It removes the step for both of you.

WebP, BMP and GIF

These are rare for documents, and they turn up anyway. A notice saved from a website may be WebP. A file from an old imaging system may be BMP. Save each as PNG. With GIF, be aware that the format allows only a small palette of colours, so a GIF of a scanned page may already be coarse. If a better source exists, ask for it.

A short conversion checklist

1. Copy the original to a safe place and work on the copy.
2. Open it and note the page count and the pixel dimensions.
3. Save or export to PNG for one page, or PDF for several.
4. Choose the highest quality or "no compression" setting offered.
5. Confirm page count and dimensions match the original.
6. Name the copy so its origin is plain: Deed packet (converted from TIFF).pdf.

Getting text out of a screenshot of a document

A screenshot of a document converts to text well when the words were large on the screen at the moment of capture. A screenshot records exactly the pixels the display showed, no more. Zoom in until the passage fills the window, capture it, save it as PNG and run OCR on that.

Screenshots are different from scans in ways that help and ways that hurt. The text is perfectly sharp, level and evenly lit, which recognition likes. But there are few pixels, and the picture includes things that are not the document.

Take a fictional matter. Vance Holloway forwards three screenshots: his loan payment history from the Northgate Credit Union member portal, a text-message exchange with a loan officer, and an email viewed on his phone. The lawyer wants the wording of each in the file as text.

The portal page. The screenshot shows a table of payments in small grey type. Ask Vance, if he can, to enlarge the page in his browser and capture it again, in sections if needed. Better still, ask whether the portal offers a download of the statement. A file exported by the portal has real text and needs no recognition at all.

The message thread. Recognition will read the bubbles as text, and it will also read the clock, the battery percentage, the contact name and the "Delivered" labels. Sender and time are evidence here, so keep them, and check that each message is attributed to the right person. Left and right alignment, which tells a human who is speaking, does not survive in plain text.

The email on a phone. The screenshot holds only what fit on the screen. A long email arrives as several overlapping captures, and the overlap gives you duplicated lines. Ask for the email itself to be forwarded. Use the screenshots only if the original is gone.

Some rules for any screenshot:

• Prefer the source. A screenshot is a picture of text that already existed as text somewhere. If the email, the web page or the file can be had, get it.
• Keep it as PNG. Screenshots are saved as PNG on most systems. Do not convert to JPG, which adds compression marks around sharp screen text.
• Dark mode is fine in principle. White text on black is still high contrast. If results are poor, try a capture in light mode.
• Watch for partial transparency and overlays. A notification banner or a floating button can sit over a line of the document.
• Do not crop away context you may need. The address bar with a URL and the timestamp may be exactly what shows where and when the content was displayed.

How screenshots should be collected and preserved for use as evidence is a legal question that varies by court and by matter. The steps here cover only how to pull the words out.

How large must the letters in an image be for OCR to read them?

Letters need enough pixels to show their shape, and the count that matters is the height of each letter in pixels. A megabyte figure or a megapixel figure says little, since a huge image with tiny text reads worse than a small image with large text. Ordinary print scanned at 300 dots per inch has enough.

The 300 figure is a common benchmark. The documentation for the open-source engine Tesseract says it works best on images with a DPI of at least 300. Other engines have their own tolerances, so treat it as a guide and test your own pages.

You can turn the benchmark into a letter height. A point is 1/72 of an inch, so type scanned at 300 dpi is about its point size times 300, divided by 72, in pixels: roughly 50 for twelve-point body text and 25 for six-point fine print. The post on scan resolution works the figures for other sizes and settings under why small print needs more dots per inch.

The same arithmetic shows why a screenshot can disappoint. A full letter-size page shown on a phone screen about 1,200 pixels wide gets roughly 140 pixels for each inch of paper. Twelve-point type is then about 23 pixels tall, less than half of what the 300 dpi scan gives it, and six-point type is down to about 12. These are estimates, good enough to compare cases.

Measuring your own image

1. Open the image at 100% zoom, so one image pixel is one screen pixel.
2. Find the smallest text you need.
3. Use the selection tool of any image editor to drag a box from the top of a tall letter such as "h" to the bottom of a descending one such as "p" on the same line. The editor shows the box height in pixels.
4. Compare it with what a 300 dpi scan would give that type: the point size times 300, divided by 72. If your figure is well under that, expect more trouble with similar-looking characters and proofread accordingly.

If the letters are too small

Enlarging the image in an editor adds pixels without adding information, so it rarely rescues text that is too small. The dependable fixes are at the source: scan again at a higher resolution, photograph from closer, or zoom the screen before taking the screenshot. The guide on how to OCR a photo of a document covers the capture technique for phones, including why standing back from the page costs so much.

Copying one passage out of an image instead of the whole page

To copy one passage out of an image, either convert the whole page and copy the lines you want from the text, or crop the image to the passage and convert only that. Converting the whole page is usually better. It keeps the passage tied to its page, and cropping too tightly can itself cause misreadings.

Route one: whole page, then copy

Upload the image, export plain text, open the .txt file and copy the passage. In Schema Law, plain text is among the exports the free plan includes. You get the entire page as text, which is useful later when someone asks what the sentence before your quotation said.

Plain text carries the words and line breaks and none of the layout. For what that means in practice, see the section on what plain text keeps and what it drops.

Route two: crop, then convert

Cropping makes sense when the page is crowded with material you do not want read: a newspaper page where you need one notice, a poster with a single paragraph of terms, a photo where the document occupies a corner of the frame.

If you crop:

• Leave a margin. Do not cut at the edge of the letters. Tesseract's documentation notes that a text area with no border at all can cause problems and suggests leaving a small border around the text. A finger's width of blank paper on each side is a sensible habit with any engine.
• Crop whole lines and whole paragraphs. A cut through the middle of a line produces half-letters that are read as something else.
• Save the crop as PNG, under a new name. Never overwrite the full image.
• Record where the crop came from. Notice photo (crop, lower left, paragraph 3).png is ugly and useful.

Then check the passage

A short passage is quick to verify and there is no excuse to skip it. Read the copied text against the image one word at a time, with attention to names, figures and any "not". If the passage will be quoted in a filing or a letter, the method in the post on how to quote text from a scanned PDF applies equally to an image.

Operating systems and phones increasingly offer a built-in way to select text in a photo. For a phone number or an address, that is handy. For anything going into a legal document, the same check applies, and you should know where the image is processed before using such a feature on client material.

Where does the text from an image file belong in the matter file?

The text from an image file belongs next to the image, under a matching name, marked as a transcription made by software. The image is the evidence of what the page or screen showed. The text is a working aid for searching, quoting and summarizing, and it has value only while someone can trace it back to its picture.

A layout that works for the Holloway matter:

A few habits keep that arrangement honest.

• Same folder, same base name. Sorting by name puts each image and its text together. Text files gathered in a separate "OCR" folder drift away from their sources within weeks.
• Put the provenance inside the text file too. File names get changed. Add a first line such as "Text recognized from Holloway - Northgate portal payments 2026-08-02 (screenshot).png. Checked against image 5 Aug 2026." and leave a blank line before the content.
• Keep converted copies with both. If the source was a TIFF or HEIC and you made a PDF or JPEG to upload, keep the original, the converted copy and the text. Label the middle one as a conversion.
• Never keep the text alone. If the image is deleted, the text becomes an unsupported assertion about what a document said.
• Quote from the text, cite to the image. The words come from the checked text file. The reference in your note or draft points to the image file and, for a multi-page source, the page.

Think about retention in both places. The copy in your matter file follows the firm's policy. The copy you uploaded to a conversion service follows that service's terms. Schema Law lets you delete a document whenever you choose, which removes the original file and everything parsed from it, and its other commitments are listed in the security section of the homepage. Once the text is exported, checked and filed beside its image, there is seldom a reason for the upload to linger.

The Holloway screenshots feed a ledger of loan payments. Put the payment date and the amount on one line, and name the screenshot beside it. If a shot is a page of the credit agreement or the promissory note, file that text apart from the payment rows. A borrower name read off a phone screen still has to be checked against the image.