Turn a Scanned Image into Text

A scanned page preserves appearance, but a picture of text is not automatically searchable text. Archive clippings, photocopied letters, signed forms, printed research, and scanner exports may remain difficult to quote or index. The Image to Text tool uses on-device OCR to create an editable plain-text draft without sending the scanned page to a third-party recognition provider.

Prepare a scan for recognition

The best scan is upright, sharply focused, and captured at enough resolution to distinguish small strokes. Large blank borders, punched holes, adjacent pages, and scanner-lid shadows can distract recognition. If possible, crop those regions before beginning. Severe rotation should also be corrected at the source; this tool does not claim a dedicated automatic deskew stage.

This preset begins with English and document preprocessing enabled. Grayscale removes irrelevant color, while adaptive thresholding responds to local background differences. It can help faded photocopies or paper with mild discoloration. On a clean, high-contrast scanner export, however, try disabling cleanup if punctuation or thin serif details appear weakened.

Choose the model matching the page language. English, Arabic, French, Spanish, German, Portuguese, and Italian are included, with a maximum of two at once. Two-language recognition is useful for parallel headings or bilingual correspondence, but it is not translation. When Arabic is selected, the editing area adopts right-to-left presentation.

What local OCR means here

Recognition begins only after you activate Extract text. The browser then loads self-hosted Tesseract.js components and the selected tessdata_fast model data. The interface reports how much uncached model data is required before loading, and Cache Storage can make that same-origin resource available for subsequent jobs. The scanned image itself stays on the device rather than being submitted to an external OCR API.

To control memory, a scan exceeding 4000 pixels on its longest edge is reduced proportionally for recognition. This happens to a temporary working copy. The source file is not changed. Named progress stages show initialization and recognition activity, and Cancel terminates the current worker when you need to stop.

From archival image to editable notes

Plain text captures words and line breaks, not the complete visual page. A table may lose columns, a marginal annotation may enter the reading order unexpectedly, and repeated page headers may need deletion. Decorative drop capitals, typewriter defects, bleed-through, stamps, handwriting, and stains all make character classification harder. Historical pages with uncommon typefaces or damaged letters deserve particularly close comparison.

Use the confidence display to focus attention. Every line can be marked low, medium, or high in addition to shading, so review does not depend on color perception. Confidence remains an estimate: a believable surname or date may still be wrong. Verify quotations, legal terms, catalogue numbers, and amounts against the scan before relying on them.

Edit the draft directly, restoring paragraphs and correcting characters. Copy is suitable for moving a passage into research notes, while TXT download creates a simple companion file for indexing or later cleanup. The original scanned image remains untouched and should be retained as the authoritative visual source.

For sensitive records, local processing reduces unnecessary exposure. A personnel page, family letter, or internal form does not have to leave the device merely to become editable. The privacy policy describes site practices beyond this tool.

Good archive work keeps provenance clear. Name the downloaded text so it can be associated with its scan, preserve uncertain readings rather than silently guessing, and record corrections when accuracy matters. OCR accelerates transcription, but the image remains the evidence. Used with that discipline, scan-to-text conversion makes a static collection dramatically easier to search and study.