Upload a clear braille photo or scan. The reader detects dot patterns, converts them to editable Unicode braille and then translates the cells into English.
High-contrast printed dots are measured as braille cells instead of guessed as English words.
Review and correct Unicode braille before back-translation.
Low-confidence photographs can use a vision-assisted fallback rather than returning a confident wrong answer.
The image reader first finds dot-shaped components and aligns them to the two-column, three-row braille grid. Each detected cell becomes a Unicode braille pattern.
Language rules are applied only after the image has been converted into cells, which keeps image recognition separate from linguistic translation.
Use side lighting so raised dots cast visible shadows. Keep the page flat, avoid glare and include a complete line without cropping its top or bottom dots.
Red or black printed dots on a plain background are currently more reliable than white embossed dots under flat frontal lighting.
It can assist, but image quality matters. AI Braille Translator uses geometric dot detection first and only falls back to vision AI when the grid cannot be resolved reliably.
The current beta accepts PNG, JPEG and WebP images up to 4 MB.
Embossed braille is supported on a best-effort basis. Side lighting and a sharp, straight photograph significantly improve detection.
Both, in sequence. The reader performs braille OCR by detecting dot cells geometrically, then applies UEB language rules to translate the detected cells into English.
Accuracy depends on image quality: flat pages, high contrast and clear dot shapes give the best results. Every detected cell remains editable before translation, so you can correct anything the reader missed.