How to Extract Text from a Photo

OCR guide · 5 min read

A photo of a page is not the same as a screenshot. When you point a phone at a printed document, a book, or a whiteboard, you introduce angle, uneven lighting, shadows, glare, curved paper, and a busy background. Optical character recognition (OCR) still works on photos, but the quality of the text you get back depends heavily on the photo you feed it. This guide walks through how to capture a good photo, prepare it, and turn it into clean, editable text you can copy anywhere.

The keyword here is extract text from a photo, and the honest answer is that half the work happens before OCR ever runs. Get the photo right and the rest is easy.

Take a photo that OCR can actually read

OCR reads the shapes of letters. Anything that distorts those shapes, blur, shadow, skew, hurts accuracy. A few habits make a large difference:

• Fill the frame. Move closer so the text occupies most of the image. A page that fills the frame gives the software far more pixels per character than one shot from across a desk.
• Shoot straight on. Hold the camera parallel to the page, directly above it, not at an angle. A slanted shot turns rectangles into trapezoids and stretches letters unevenly.
• Get even, indirect light. Soft daylight near a window works well. Avoid a single harsh lamp or direct flash, which creates a bright hotspot (glare) that erases whole lines.
• Tap to focus. On a phone, tap the text before shooting so it locks focus there. Sharp edges matter more than resolution.
• Flatten the page. For books and glossy magazines, press the page flat or weigh down the corners. Curved paper bends text near the spine, where OCR struggles most.

If you can retake the shot, retake it. Two seconds of better lighting saves you minutes of correcting garbled text later.

Why "Accurate" mode matters for photos

Clean screenshots have crisp, high-contrast pixels, so a fast recognition pass reads them almost perfectly. Photos are messier, so they benefit from a slower, more careful model. Textquill offers two modes for exactly this reason: a Fast mode for clean, digital-native images and an Accurate mode that spends more effort resolving ambiguous characters in photographs. There's also an Auto setting that inspects each image and picks the mode for you, which is a sensible default when you're processing a mix of screenshots and real-world photos.

For a photo of a receipt, a book page, or a whiteboard, choose Accurate if the automatic pick looks shaky. The extra second or two of processing is usually worth it. All of this runs on your device, so nothing is uploaded and the photo never leaves your computer.

Prepare the image before you run OCR

A little editing before recognition removes distractions that confuse the software.

Crop to the text

Crop away the desk, your fingers, the table edge, and anything that isn't the text you want. A tighter crop means the OCR engine isn't wasting effort trying to interpret background clutter as characters.

Rotate and straighten

If the photo came out sideways or tilted, rotate it so lines of text run horizontally. Most OCR expects roughly level baselines; a page rotated even 15 degrees can drop accuracy noticeably.

Handle multiple columns

Newspapers, academic papers, and some product manuals use two or three columns. OCR can read them, but it sometimes stitches lines across columns in the wrong order. If the output looks scrambled, crop and process one column at a time, then paste the results together. It's a small extra step that keeps the reading order intact.

A realistic workflow

Say you photographed a page from a book to quote it later. Here's a practical sequence:

1. Photograph the page straight on, near a window, with the page pressed flat and the frame filled.
2. Open Textquill and drag the photo in, or paste it with Ctrl+V (Cmd+V on a Mac). You can also upload it from disk.
3. Crop to the paragraph you need and rotate if the shot was tilted.
4. Set the mode to Accurate (or leave it on Auto) and run the extraction.
5. Read the result, fix any obvious slips, then copy the text or export it as TXT or Markdown.

The same routine works for a business card, a product label, a street sign, or a whiteboard after a meeting, adjust only the crop and the lighting to match.

Clean up the recognized text

Even a good photo produces small errors, and knowing where to look saves time:

• Look-alike characters. Scan for common swaps: the number 0 versus the letter O, 1 versus lowercase l, and rn misread as m.
• Broken line breaks. Photos of justified text sometimes add stray line breaks mid-sentence. Rejoin paragraphs so the text flows.
• Spacing near punctuation. Check for missing spaces after periods and commas, especially on low-contrast prints.
• Language settings. If you photographed text in French, German, or another language, make sure the recognition language matches. Textquill supports 16 languages, and picking the right one improves accented characters.

Once the text is clean, you can use Textquill's read-aloud feature to hear it back, a quick way to catch errors your eyes skip over, and keep a copy in history if you'll need it again.

FAQ