Tamil OCR

1. Choose your source. Drop a screenshot or scan, choose a PDF, paste an image from the clipboard, or use Camera on your phone. The file name appears above the options. Selecting a file does not start recognition. To see how it works first, press Try a sample image, which loads our sample and reads it straight away.

2. Set the language and orientation. Tamil is the default. Choose Tamil + English for mixed pages. If the scan is sideways, select a clockwise rotation. The same rotation applies to every page of a PDF, so separate pages with different orientations first.

3. Press Read text. The first run downloads the recognition engine and the selected language model from this site, once: about 7.2 MB for Tamil with High Accuracy (the default), or 4.5 MB with Fast. Tamil + English adds 12.8 MB with High Accuracy or 2 MB with Fast, and your first PDF adds about 1.5 MB for the PDF reader. Your browser keeps these files, so later runs start much sooner. Progress shows loading and recognition separately. PDFs are processed one page at a time. Keep the tab open; a long document can take several minutes.

4. Review before you reuse. Edit the text directly. Highlighted words below the editor have low recognition confidence; click one to select its first matching occurrence. Check repeated words against the page number in the button’s tooltip. Copy the result or use More to download a text or Word file.

The output contains text, not a reproduction of the original page. Columns, tables, headings, pictures and page layout are not preserved in the downloaded Word document. Keep line breaks if the source is poetry or a list. Join Wrapped Lines combines adjacent lines while preserving blank lines and the gaps between PDF pages. Choose this option before recognition.

On four short synthetic Tamil images, the default High Accuracy model had 0% character error after Unicode cleanup (11.3% before cleanup); this tiny test does not predict accuracy on your scans.

High Accuracy is the default because it read our small test set more accurately; Fast is available for comparison. Both use Tesseract language models. The second model downloads only when you run recognition with it selected. Try both on a representative page before processing a long document. The name High Accuracy describes the model family, not a promise that it will read every image better.

The tool normalises Unicode vowel signs such as கொ into கொ and removes stray invisible joiners. It deliberately avoids automatic spelling changes. A confident error is still an error: check names, dates, amounts and unfamiliar words carefully. For essential records, compare the complete result with the scan, rather than reviewing only highlighted words.

Legacy-font text in a PDF can look like Tamil while copying as Latin characters. If you already have that extracted text, try the Tamil font converter first. OCR is useful when the page is an image or when ordinary copying cannot recover readable text.

Start with a flat, evenly lit page. Hold the camera parallel to the paper and avoid shadows, glare and motion blur. Crop out the desk and neighbouring pages before choosing the file. Keep a small clear margin around the text; a tight crop can cut off vowel signs. Scanning at about 300 dpi is a useful starting point recommended by Tesseract.

Use dark text on a light background. A sharp scan usually works better than a heavily compressed screenshot. For small text or uneven lighting, try Auto-Enhance. It enlarges small images, removes colour and compares each pixel with its neighbourhood to separate ink from paper. This can also erase thin strokes, so leave it off for clean originals and compare the results on difficult photos.

Printed Tamil is the intended input. Handwriting, highly decorative fonts, old palm-leaf scripts, curved book pages and damaged letters may produce little usable text. For multiple columns or dense tables, crop and recognise each region separately to control reading order. This tool cannot reconstruct a missing or illegible character reliably.