When you photograph a page, your phone stores a picture of the text, not the text itself. You cannot search it, select a sentence or paste it into an email. Optical character recognition, or OCR, is the technology that reads the picture and produces real text. It is why you can search a folder of scanned receipts or copy a paragraph from a photographed notice.
How OCR works, in plain language
Modern OCR engines work in stages. First they clean the image: straighten it, adjust contrast and separate text from background. Next they find blocks of text, then lines, then words and characters. Finally a recognition model decides which letters it is looking at, using context from the language to resolve ambiguity, for example distinguishing the digit 0 from the letter O in the middle of a word.
The output is text plus information about where each word sits on the page. That position information is what makes a searchable PDF possible.
What a searchable PDF actually is
A searchable PDF keeps the original picture and places an invisible layer of text exactly over the words. To the eye it is identical to the scan, but when you search, select or copy, the viewer uses the hidden text. The picture is not changed or replaced, so the page still looks like the original and signatures and stamps are preserved.
That is how the Image to PDF tool works: it embeds your pictures at their original quality and adds the recognised words underneath, so the finished PDF can be searched and copied from.
What affects accuracy
- Resolution. Aim for the text to be at least 20 pixels tall. For a normal document that means roughly 200 to 300 dpi when scanning, or holding a phone camera close enough that the page fills the frame.
- Lighting and shadows. Even, bright light without a shadow across the page matters more than a high-megapixel camera.
- Straightness. Photograph from directly above. Keystone distortion and curved book pages reduce accuracy.
- Contrast. Dark text on a light plain background is ideal. Coloured backgrounds, watermarks and textured paper make recognition harder.
- Font and print quality. Clean printed text is easy; faded dot-matrix printing, tiny fonts and decorative typefaces are harder.
- Language. The engine must know which language to expect. Choosing the right language avoids garbled output.
Handwriting
Printed text is largely a solved problem; handwriting is not. General-purpose engines usually fail on cursive and are unreliable on neat handwriting. If you need to digitise handwritten notes, expect to retype them or use a specialist handwriting tool and proofread the results.
Indian languages and other scripts
OCR works for many scripts, including Devanagari (Hindi, Marathi), Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Gurmukhi (Punjabi), Odia and Urdu, as long as language data for that script is installed. Accuracy is generally lower than for English because these scripts have joined letters and many conjunct characters, so clean, high-resolution input matters even more. Pick the document's language in the tool; if the page mixes English with another language, English is added automatically.
A checklist for good results
- Place the page flat on a plain surface in daylight or under a lamp that does not cast a shadow.
- Fill the frame with the page and keep the phone parallel to it.
- Take a sharp picture: tap to focus, hold still.
- Avoid filters, beauty modes and heavy compression.
- Choose the correct text language before creating the PDF.
- Open the PDF, search for a word you know is present, and copy a sentence to check accuracy.
- Proofread important numbers such as amounts, dates and ID numbers. OCR can confuse similar characters.
Privacy considerations
Documents you scan often contain personal details. Prefer tools that process the file and hand it back without storing it, avoid uploading documents you do not need to, and delete local copies you no longer require. TinyFileHub does not keep your files after delivery; read our privacy policy for details.
When OCR is not the right tool
If you already have the original digital file, a word-processor document or a text-based PDF, use that; OCR only guesses what a person could read. For forms and tables, expect to check the layout manually, because OCR extracts words, not always the structure of columns. And for very poor scans, re-scanning is often faster than correcting recognition errors.
Used sensibly, OCR turns piles of photographed paper into searchable, copyable documents in seconds. Good input is most of the work: a flat, bright, sharp page in the right language gives results that need little correction.
Everyday uses
- Searching old paperwork. Receipts, warranty cards and bills become findable by typing a shop name or amount.
- Copying from photographs. A notice board, slide or printed table in a photo can be pasted into an email.
- Studying. Photographed textbook pages become searchable so you can jump straight to a term.
- Translating. Copy the recognised text into a translator instead of retyping it.
- Accessibility. Real text can be read aloud by screen readers.
When the text layer is wrong
Because the picture is untouched, a recognition error never changes how the document looks; it only affects what you copy or search. Typical errors are 1 and l, 0 and O, rn and m, and missing diacritics. Before relying on numbers copied from a scan, compare them with the picture. For tables, expect the words to come out in reading order but the column structure to be lost.