How to Extract Text From an Image With OCR
OCR turns a picture of words into text you can copy and edit. Here's how it works and the simple fixes that make results far more accurate.

On this page
You've got a photo of a whiteboard after a meeting, a screenshot of an error message, or a scanned recipe from your grandmother's cookbook. The words are right there in the picture, but you can't copy them, search them or paste them into an email. Retyping is slow and you'll introduce typos.
That's the job OCR (optical character recognition) was built for. It reads the text in an image and hands it back as real, editable characters. Here's how it works in plain terms, how to get clean results, and where its limits are.
What is OCR and how does it work?
To a computer, an image is just a grid of colored dots. It has no idea that a particular cluster of dark pixels is the letter "a." OCR software bridges that gap in a few stages.
- Clean-up. The image is converted to high-contrast black and white, straightened if it's tilted, and cleared of specks.
- Layout detection. The engine finds where the text is: separating blocks, columns, lines and individual words from pictures and blank space.
- Character recognition. Each word or character is compared against patterns the engine has learned. Modern engines use neural networks trained on huge amounts of text, so they recognize whole words in context rather than letter by letter.
- Correction. Language models check the output against real words. If the shapes could be "c1ear" or "clear," the dictionary nudges it toward "clear."
The output is plain text you can copy, edit and search.
How do you extract text from an image?
The Image to Text tool on ImgPress runs OCR right in your browser. Your image isn't uploaded anywhere, which matters when it's a contract, a bank letter or anything else you'd rather keep private.
- Open Image to Text.
- Drop in your image, or paste a screenshot straight from your clipboard.
- Choose the language of the text.
- Run the recognition and wait a few seconds.
- Review the output, fix any errors and copy it.
The first run may take a bit longer while the language data loads. After that, it's usually quick.
How do you get accurate OCR results?
OCR quality depends far more on the image than on the software. A sharp, straight, well-lit photo of printed text can come out nearly perfect. A blurry, angled shot under a yellow lamp will produce gibberish no matter which engine you use.
| Problem | What it causes | Fix |
|---|---|---|
| Low resolution or tiny text | Letters merge, l/1/I mix-ups | Photograph closer; aim for text at least 20 to 30 pixels tall |
| Blur or camera shake | Missing or garbled words | Tap to focus, hold steady, use good light |
| Tilted or curved page | Lines get split or merged | Shoot straight on; flatten the page; rotate the image level |
| Low contrast (grey on grey) | Characters not detected | Increase contrast; use even lighting |
| Shadows across the page | Patches of missing text | Move to diffuse light, such as near a window |
| Busy backgrounds or extra content | Stray characters, noise | Crop to just the text area |
| Wrong language selected | Accented letters wrong, misread words | Pick the correct language before running |
Crop to what you need
If you only want one paragraph from a full page, or the address from a parcel label, crop to that area first with Crop Image. Less clutter means fewer stray characters and a faster result.
Get the text level
OCR engines can correct slight tilts, but anything more than a few degrees causes trouble. If the photo is sideways or skewed, fix it with Rotate Image before you start.
Improve weak images
Faded receipts and dim photos often benefit from a quick boost to contrast and sharpness. The Image Enhancer can help, though it can't rescue text that's genuinely unreadable to begin with.
Tip: If one block of text keeps coming out wrong, crop just that block and run it again. Engines often do better when they don't have to work out a complicated layout.
Can OCR read handwriting?
Sometimes, and it depends heavily on the writing. Neat, printed capitals (like on a form) can come out well. Joined-up cursive, doctors' notes and quick scribbles are much harder, because letters vary a lot between people and even within one person's writing.
Standard OCR engines are mostly trained on printed text, so expect lower accuracy on handwriting and plan to proofread every line. Specialized handwriting recognition exists and does better, but no system reads every hand reliably. For a few lines of handwriting, typing them yourself is sometimes faster than correcting the output.
Does OCR work on screenshots?
Very well, usually better than on photos. Screenshots are already perfectly straight, evenly lit and high contrast, which is exactly what OCR likes.
Common uses:
- Copying an error message so you can search for it
- Grabbing text from a video frame, a slide in a meeting or a PDF that won't let you select text
- Pulling quotes from an image posted on social media
- Capturing a list from an app that blocks copying
Two things can trip it up: very small UI text at low zoom, and stylized fonts or text over busy images. If the result is poor, zoom in before taking the screenshot so the text is larger.
Note: Your phone may already do this. iPhones have Live Text built into the camera and Photos app, and many Android phones can do the same through Google Lens. For desktop screenshots or batch work, a dedicated tool is often more convenient.
Is it safe to use online OCR tools?
That depends on where the processing happens. Many online OCR services upload your image to their servers, run it there and send back the text. For a cookbook page, that's fine. For a passport, a medical letter, a payslip or a client contract, it's worth pausing.
Questions to ask:
- Is the image uploaded, or processed on my device?
- How long is it kept, and is it used to train anything?
- Does the service need an account, and what's linked to it?
Browser-based tools that run OCR locally avoid the issue entirely: the image never leaves your computer or phone. That's how ImgPress works for every tool, OCR included.
What can you do with the extracted text?
Once the text is out, it behaves like anything you've typed. A few practical ideas:
- Make scanned documents useful. Pull the key details from a letter into your notes or a spreadsheet.
- Translate. Paste extracted text into a translator when you're travelling or dealing with foreign-language paperwork.
- Search your archive. Add OCR text to file descriptions so you can find "that receipt from the plumber" later.
- Accessibility. Extracted text can be read aloud by a screen reader, unlike text trapped inside an image. When you're publishing images online, describing them properly helps too; see our image alt text guide.
Always proofread the output
Even excellent OCR makes small mistakes: a 5 that should be an S, a missing decimal point, two words joined into one. In a casual note, that's harmless. In an invoice amount, a phone number or a legal clause, it isn't.
Read through the result, check every number against the original, and keep the source image until you're sure the text is right. OCR saves you most of the typing. The final check is still your job.
Written by ImgPress Editorial
Our editorial team writes and maintains practical guides on working with images. Articles are reviewed for accuracy and updated when tools or best practices change. Spotted an error? Let us know.
Questions readers ask
Can OCR read text in any language?
OCR engines support many languages, but each one has to be trained for a specific script and vocabulary. Selecting the correct language before you run it noticeably improves accuracy, especially for accented characters and non-Latin scripts.
Why does OCR confuse letters like l, 1 and I?
Some characters look nearly identical in many fonts, particularly at small sizes or low resolution. The engine relies on shape and context to decide, so blurry or tiny text makes these mix-ups more likely. Higher resolution and sharper images help.
Does OCR keep the formatting of the original?
Basic OCR usually returns plain text with line breaks, not fonts, colors or exact layout. Tables and multi-column pages may need tidying after you paste the text into a document.


