How to Take Good Photos of Documents for OCR

OCR works great on clean inputs and miserably on bad ones. The difference between “99% accuracy” and “I can’t even read what came back” is usually the photo, not the software.

Here’s what actually matters when phone-photographing paper documents for OCR — and the tricks that get you the cleanest possible result.

The basics that matter most

In rough order of impact:

1. Lighting: even, indirect daylight is ideal. Avoid:

Direct overhead lights — they create reflections off the paper
One-sided lighting — half the page is bright, half dim
Your shadow falling on the page
Glossy paper under bright light — gloss reflects

If you’re indoors, position the document near a window in soft daylight. If artificial light is the only option, use a flat workspace away from direct lamp glare, ideally with multiple light sources to balance any shadow.

2. Angle: hold the phone directly above the page, with the camera plane parallel to the paper. Tilted angles cause keystone distortion — the top of the page renders smaller than the bottom (or vice versa). OCR struggles with text whose size shifts across the page.

A good test: when you frame the document in your viewfinder, the corners should form a near-perfect rectangle. If it’s trapezoid-shaped, adjust your angle.

3. Distance and framing: fill the frame with just the document. Don’t include lots of background table, your hands, surrounding objects. Tight framing means more pixels-per-character, which OCR loves. But don’t get so close that focus becomes a problem.

4. Focus: tap the page in your camera app before shooting. Most phones default to auto-focusing on whatever’s in the center; tapping locks focus to where you want it. For documents specifically, tap a text-rich area.

5. Background: a dark, plain background (a black table, a black sheet of paper underneath) helps any automatic page-detection in scanner apps. White-on-white (document on white desk) makes edge detection harder.

What “good enough” actually looks like

A few specific quality thresholds:

Resolution at the page: aim for ~300 DPI equivalent. A standard 8.5×11” page photographed to fill an iPhone’s full 4032×3024 sensor gives you roughly 366 DPI — plenty. As long as the page fills most of the frame, modern phone cameras have enough resolution.
Sharpness: text should be crisp. Zoom into your photo before saving — can you read individual letters clearly? If letters look soft or doubled, you’ve got camera shake or focus issues. Reshoot.
Contrast: black text on white paper. A well-lit photo gives you near-black ink on near-white paper. If your photo shows gray text on cream paper, your OCR will struggle — the recognizer needs clear edge contrast.
No motion blur: handheld photos at long exposures (low light) produce subtle motion blur that humans tolerate but OCR doesn’t. In dim conditions, brace your hand against something solid or use a timer.

Use your phone’s scan mode when possible

Most modern phones have a dedicated “scan documents” mode hidden in a standard app:

iPhone: open Notes → tap the camera icon → “Scan Documents.” Or open Files → tap the three-dots menu → “Scan Documents.”

Android (Google ecosystem): open Google Drive → tap the ”+” → “Scan.”

Samsung phones: open Camera → look for “Document” or “Scan” mode in the modes menu.

These built-in scan modes do four things that improve OCR dramatically:

Auto-detect the page edges (you don’t have to crop)
Auto-flatten any keystone distortion from your angle
Enhance contrast to make text crisper
Multi-page support — keep adding pages to build a multi-page PDF

If your phone has a scan mode, use it. The output is significantly better than regular camera mode for OCR purposes.

When you have to use regular camera

If you don’t have a scan mode (older phone, taking photos in a context where opening Notes is awkward), regular camera works but needs more care:

Set the camera to a high-resolution mode (not Live Photo if your phone has the option; not Portrait mode either — both add weirdness OCR doesn’t like)
Hold the phone parallel to the page, directly above
Tap to focus on the text before shooting
Take 2-3 photos of each page — one might come out sharper than another

After capture, run the photos through the Image Cropper to trim to just the page area before OCR.

The “magic” enhancements that don’t actually help

A few things people think improve OCR but don’t:

Turning the photo grayscale yourself. OCR engines do this internally; pre-converting saves no time and loses information that might help recognize accents in non-English documents.

Boosting contrast aggressively. Tesseract handles normal contrast just fine. Aggressive contrast pushing can blow out fine details (the bottom of “a” closing up, the gap in “e” filling in) and hurt accuracy.

Increasing photo resolution beyond ~300 DPI equivalent. OCR doesn’t get better past about 300 DPI for printed text. Higher resolution just makes files bigger.

Using a “scan filter” in photo apps (the Instagram-style “old paper” filter or whitening filters). These add artifacts that confuse OCR. Skip them.

What to do about reflections

Glossy paper (some magazines, photo prints, badly-laminated documents) reflects light, washing out portions of the page. A few workarounds:

Angle yourself to put reflections off-page rather than on the text
Use diffused lighting — a single light source bounced off a wall creates softer reflections than direct light
For very reflective documents, scan the document in shadow (next to a window but with the page in indirect light)

If a reflection is unavoidable, the scanner app’s “enhance” mode sometimes removes them automatically. Worth trying before resigning yourself to a reshoot.

What to do about wrinkled or curled paper

Pages that bend away from the camera (curled at the edges, folded from being in a wallet) confuse both page-detection and OCR. Flatten the paper:

Place it under a clear glass sheet (a picture frame works) to hold it flat
Or place a book on top for a few hours before scanning to flatten naturally
For chronically curled paper (rolled receipts), iron on low heat through a cloth (yes, really — this works) or just bend the page the opposite direction by hand briefly

Multi-page documents

For multi-page paper documents (booklets, contracts, books):

Phone scan modes handle this directly — keep adding pages, end up with a multi-page PDF.

Without scan mode: photograph each page separately, then bundle:

Crop each photo with Image Cropper
Combine into a PDF with Images to PDF
Run the combined PDF through PDF OCR

Tedious for many pages — phone scan mode is much faster if available.

Special-case advice

Receipts (long, narrow, often thermal paper):

Lay flat (uncurled)
Photograph at a slight zoom-in — receipts have small print and need pixels-per-character
Run individually through Image to Text rather than bundling many into a PDF

Books:

Press the binding down firmly so the page is flat
Photograph each page separately (or use the phone scan mode that handles two pages at once)
Be patient with thick books — gutter (binding curvature) hurts OCR near the spine

Forms with handwriting:

OCR handles printed text well, handwriting poorly
For mostly-printed forms with a few handwritten fields, OCR gets you the structure; you’ll need to manually fill in the handwritten parts

Business cards:

Bright, even lighting essential — cards are small, every pixel matters
A single card is best photographed in regular camera mode (filling the frame), not scan mode (which expects page-sized inputs)

Old / faded documents:

Beyond what photo technique can fix
Try to increase contrast in the source document (photocopy at higher contrast before OCR) rather than relying on photo capture to recover

The full workflow

Putting it together:

Photograph each page with phone scan mode (best) or regular camera (acceptable)
Inspect each photo before moving on — sharp? Well-lit? Not skewed?
Crop with Image Cropper if needed (scan mode usually handles this)
Bundle into a PDF if multi-page with Images to PDF
OCR with PDF OCR (multi-page) or Image to Text (single image)
Spot-check the recognized text — is it accurate? If not, reshoot.

The reshoot loop is the secret to high-quality OCR results. Don’t accept poor recognition — figure out what was wrong with the photo and try again.

Privacy

Browser-based OCR tools mean documents never get uploaded. For photographing private documents (medical paperwork, financial records, contracts), this matters. The photos can live entirely on your phone and laptop; the OCR happens locally.

TL;DR

Light evenly, hold the phone parallel to the page, fill the frame
Use your phone’s scan mode if it has one — iPhone Notes, Google Drive on Android
Tap to focus on text, take 2-3 photos per page
Crop to just the page with Image Cropper if needed
Bundle multi-page with Images to PDF
OCR with PDF OCR or Image to Text
Reshoot if accuracy is bad — bad photos can’t be salvaged by better OCR software