OCR Test

Experimental app for optical character recognition (OCR)

This app is an experimental app that I developed several years ago that demonstrates use of the Tesseract OCR engine to recognize text in images captured by the device camera.

This app runs OCR on your device – without uploading your images to a server – and is suitable for recognizing individual words or short phrases of text, but this app is intended for hobbyists and software developers interested in OCR and not for general audiences.

In contrast to Google's Mobile Vision API, this app is able to recognize text printed in non-Latin-based fonts while offline. To achieve this, this app incorporates an unusually large amount of training data for several languages. This training data is stored on your phone, and this app takes up much more space than ordinary apps.

No image pre-processing is performed by this app before handing off captured image frames to Tesseract, so the app is not tuned for any specific use case and, as a result, its recognition accuracy and speed is heavily dependent on situational factors like perspective, lighting, and font type.

Source code for this app is available on GitHub (with minor changes to accommodate GitHub file size restrictions). The code for this app is a combination of open source camera-related code from the ZXing bar code scanner project and open source optical character recognition code from the Tesseract OCR project.

TEXT CAPTURE

The default single-shot capture runs OCR on a snapshot image that's captured when you click the shutter button, like a regular photo.

When the "continuous preview" checkbox is checked, the app shows a dynamic, real-time display of what the device is recognizing right beside the camera viewfinder. The continuous preview mode works best on a fast device.

USING THIS APP

• Point the device at a small region of text and touch the on-screen shutter button to start OCR.

• For recognizing individual Chinese/Japanese/Korean characters, set the page segmentation mode to "single character."

RECOGNITION ACCURACY

• Various factors can cause the OCR to fail: uneven illumination, stylized text, or text without enough contrast from the background. Try to have good lighting.

• Hold the device steady, and be sure the picture is in focus.

• If you need to scan a large block of text or an entire document, try a document scanning app such as Text Fairy instead.

LANGUAGES

• This app supports several languages/scripts not recognized by Google Translate.

• Supported languages for OCR:

Afrikaans
Albanian
Amharic
Arabic
Assamese
Azerbaijani
Azerbaijani (Cyrillic)
Basque
Belarusian
Bengali
Bosnian
Bulgarian
Burmese
Catalan
Cebuano
Cherokee
Chinese (Simplified)
Chinese (Traditional)
Croatian
Czech
Danish
Dutch
Dzongkha
English
English, Middle (1100-1500)
Esperanto
Estonian
Finnish
Frankish
French
French, Middle (ca. 1400-1600)
Galician
Georgian
Georgian - Old
German
Greek, Ancient (-1453)
Greek, Modern (1453-)
Gujarati
Haitian
Hebrew
Hindi
Hungarian
Icelandic
Indonesian
Inuktitut
Irish
Italian
Italian - Old
Japanese
Javanese
Kannada
Kazakh
Khmer
Korean
Kurdish
Kyrgyz
Lao
Latin
Latvian
Lithuanian
Macedonian
Malay
Malayalam
Maltese
Marathi
Nepali
Norwegian
Oriya
Pashto
Persian
Polish
Portuguese
Punjabi
Romanian
Russian
Sanskrit
Serbian
Serbian (Latin)
Sinhala
Slovak
Slovenian
Spanish
Spanish - Old
Swahili
Swedish
Syriac
Tagalog
Tajik
Tamil
Telugu
Thai
Tibetan
Tigrinya
Turkish
Ukrainian
Urdu
Uyghur
Uzbek
Uzbek (Cyrillic)
Vietnamese
Welsh
Yiddish

SAMSUNG DEVICE NOTES

• On Samsung Galaxy devices, you may need to long-press the menu button to set preferences.

• You may get better results if you un-check "Standard focus mode".