OCR accuracy tips for phone scans and screenshots

Optical character recognition guesses letters from shapes. It does not understand meaning. A “0” and an “O” look alike when the photo is blurry. A comma becomes a period when contrast is low. Browser OCR on QuickConvert uses Tesseract locally via Image OCR, which keeps scans private but still rewards good inputs.

Noise from JPEG compression, motion blur, and skewed paper are the usual culprits. Fixing the image before OCR is cheaper than fixing a pasted table afterward.

Start with light and geometry

Photograph documents straight-on under even light. Avoid ceiling spotlights that glare on glossy paper. If the page curves in a booklet, flatten it with weights or scan one side at a time. Rotate so text lines run horizontally; the engine expects landscape reading order.

Crop with Image Crop so margins, keyboards, and thumbs disappear. Less clutter means fewer false characters at the edges.

More pixels help only until characters are sharp. A 300 dpi flatbed scan beats a 48 MP glare photo. If a phone shot is sharp, OCR may downscale internally anyway. When letters look soft on screen, move closer or tap to focus before you shoot.

Screenshots of PDFs or slides are excellent sources when saved as PNG. Export a stubborn page with PDF to Images if the PDF viewer will not let you copy text.

Colour and contrast

Black text on white paper is ideal. Coloured stationery and yellow highlighter confuse binarization. For coloured backgrounds, convert to grayscale in a photo app or increase contrast before OCR. Inverted white-on-black slides sometimes OCR better after you invert back to black-on-white.

Pick the dominant language on the page before you run Image OCR. Mixed Arabic and English forms may need two passes on cropped regions. Multi-column layouts read across unless you crop each column. Tables rarely rebuild perfectly; expect to fix column alignment in a spreadsheet.

Numbers and punctuation

Amounts and dates are high-stakes. Read every digit after OCR. Thousand separators and currency symbols are frequent errors. For receipts, compare the OCR total to the picture, not to memory.

Legal citations and footnotes in tiny type need crop zoom. Run OCR on the main body first, then crop footnotes separately.

Image Compress can help email-sized files but harsh JPEG can erase serifs. If compression makes strokes fat, OCR confuses “rn” with “m”. Prefer a mild compress pass or crop instead of crushing quality.

Phone “Live” photos sometimes sharpen worse than a still. Use the still frame.

Privacy-sensitive scans should stay local. After OCR, paste into a draft, or rebuild a PDF with Office to PDF once text is clean. Merge exhibits with PDF Merge when the packet must be one file.

When a long batch scan must be split by document, read the guide on separating scanned PDFs, export each section, and OCR per piece.

When not to trust OCR alone

Handwriting, stamps, and faint faxes need humans. Signatures are not reliably captured. For regulatory filings, treat OCR as a draft. Store the image as evidence even after you paste text.

A $50 LED panel behind the camera reduces shadow on phone captures. Scanner apps with edge detection help, but always open the result before OCR. Name files by date and vendor so batches sort predictably.

Re-run OCR after small edits. A five-degree desk skew might not look ugly to you but can cost five percent accuracy on small print.

Should I OCR the whole PDF or each page? Export pages as images when the PDF has no text layer. OCR each page or crop regions. Whole-PDF server tools upload; browser workflow is per image.

Does sharpening help? Mild sharpening can help soft scans. Aggressive sharpening invents edges OCR reads as punctuation. Prefer a second photo with better focus.

Can OCR read Arabic diacritics? Quality varies by font and scan. Proofread names and religious text carefully. Crop to a single column when mixing scripts.

Will OCR fix a rotated PDF? Rotate the image before OCR. The tool does not auto-deskew. Some scanner apps deskew on save.

Is cloud OCR more accurate? Sometimes, but it uploads your document. For personal paperwork, local OCR plus careful prep is the safer default on QuickConvert.

What about spreadsheets photographed at an angle? Expect grid chaos. Retake or use a scanner. OCR is for prose blocks, not reconstructing Excel.

Should I delete the image after OCR? Keep the image until the text is verified and archived. OCR output is not a legal archive format.

Does dark mode UI in screenshots hurt OCR? Light mode exports often OCR better. Invert colours if you must process a dark screenshot.

Publish a short checklist for anyone photographing paperwork: level the page, fill the frame, crop before OCR, pick language, proof numbers twice. Teams that share intake scans reduce support tickets when every photo looks alike. Store the checklist beside the link to Image OCR in your wiki.

When a project spans weeks, rename OCR exports with dates so you can match text files to pictures later. A plain folder of IMG_2045.jpg without notes forces redo work. Browser tools do not organise archives for you; naming discipline does.

Revisit samples monthly: one new phone model or OS update can change default HDR processing and silently hurt OCR until you notice.