Image to Text — OCR Converter

What This Tool Does

This OCR tool reads the pixels of an image and converts recognizable text characters into a plain-text string you can copy, search, and edit. It handles printed documents, screenshots, photos of signs, book pages, receipts, business cards, and more.

The OCR engine is a WebAssembly build of a battle-tested open-source recognition pipeline. It runs entirely in your browser, so image data never leaves your device.

How Does It Work?

OCR works in a pipeline of image processing stages before any character recognition begins:

Image input
  → Grayscale conversion
  → Binarization (Otsu thresholding: separate text pixels from background)
  → Deskew (detect and correct page tilt up to ±15°)
  → Layout analysis (detect paragraphs, columns, text blocks)
  → Line segmentation (split into individual text lines)
  → Word segmentation (split lines into words)
  → Character classification (LSTM neural network → Unicode codepoints)
  → Language model post-processing (word-frequency scoring)
  → Plain text output

Language models are loaded on-demand. Selecting “English” loads an ~8 MB LSTM model trained on printed English text. Other languages load separate model files.

How Does DPI Affect Image Quality?

Image resolution is the single biggest factor in OCR accuracy:

Source	Typical DPI	Expected Accuracy
Phone photo (well-lit)	200–300	92–97%
Scanner (document mode)	300	96–99%
Screenshot (Retina 2x)	144–288	95–98%
Screenshot (1x)	72–96	85–93%
Scanned fax	100–150	75–88%

For printed documents, 300 DPI is the practical minimum for reliable accuracy. Scanning at 600 DPI rarely improves accuracy further but increases processing time.

What Are Common Use Cases?

Digitizing receipts and invoices. Photograph a paper receipt and extract the line items and totals to paste into a spreadsheet — much faster than manual retyping, especially for expense reports.

Copying text from screenshots. When a website, PDF, or app doesn’t allow text selection, take a screenshot and run it through OCR to get selectable, searchable text.

Extracting quotes from books. Photograph a page and get the exact quote text in seconds instead of transcribing it by hand. Useful for researchers, students, and writers.

Making scanned PDFs searchable. Extract the text from each page image, then use the text to create searchable content or to index the document.

Reading text in images for accessibility. Convert image-based text to plain text so it can be read aloud by screen readers or translated by language tools.

Frequently Asked Questions

Which image formats are supported? JPG, PNG, WebP, TIFF, and BMP. PNG and TIFF are lossless formats that preserve text sharpness, which is why scanned documents saved as PNG generally produce better OCR results than JPEG exports from the same scan. Avoid JPEG quality settings below 80 for documents.

Can it read handwriting? Printed text is where OCR excels. The engine has limited support for neat, well-spaced handwriting — especially block letters — but cursive and informal handwriting produce unreliable results. For important handwritten documents, treat the OCR output as a first draft and expect significant manual correction.

My image has text in multiple columns. Will it read them in order? The layout analysis step detects obvious two-column layouts, but complex magazine or academic paper layouts with multiple interleaved columns, sidebars, and captions often confuse the reading-order logic. For these documents, extract the text and rearrange paragraphs manually.

Does the tool preserve formatting like bold or tables? No. The output is pure plain text — no styling, no table grid, no font attributes. If you need table data, the numbers will appear in the output but without column alignment. Recreating the table structure in Excel or Google Sheets requires manual arrangement.

Why is some text missing from the output? Common causes: resolution too low (under 100 DPI effective), image is blurry or out of focus, text is printed on a dark background without sufficient contrast, or the font is highly decorative. Try increasing the image contrast and brightness before running OCR again.

Is the tool free for commercial use? Yes. There are no usage limits and no account required. The underlying OCR engine is released under a permissive open-source license, which permits commercial use without royalties or attribution requirements.