OCR

OCR is an Android front end for Tesseract 5. It extracts text from an image chosen from the crop screen, captured with the system camera, or shared from another app, and keeps a local history. The history screen itself has no text search. The project lists more than 120 recognition languages.

Version 6.1 rebuilt the interface in Jetpack Compose and added a Room-based history, an accuracy display, summary cards and language filtering. Sharing an image into the app already existed before this release. Version 6.1 also raises the minimum to Android 6.0.

Models are downloaded separately

The 35.0 MB installer contains the OCR engine and native libraries for four CPU types, but it does not contain English or any other Tesseract language model. No language starts selected. Most choices download .traineddata files over HTTPS from the official tesseract-ocr/tessdata_best repository on GitHub. Akkadian instead comes from tessdata_contrib, while the equation model uses the older tessdata 3.04.00 file.

After a model is present, recognition runs locally through Tesseract4Android. The crop interface can open the gallery or launch the phone’s system camera app, so OCR itself does not request camera permission. Android’s share sheet is a third input route. The app’s manifest asks only for internet, network-state and an internal AndroidX receiver permission. Our static scan found no known tracker SDKs.

The downloader does not compare the file with a hard-coded checksum in version 6.1. HTTPS protects the transfer in transit, but there is no second model-hash check in this app. An interrupted download can leave a partial file that the app later treats as an installed model. Reinstalling or adding another language requires network access again.

History keeps an image copy

There is an important storage detail that is not obvious from the feature list. Our static review of the fixed 6.1 source shows that OCR results—including failed recognitions—are saved in Room and retain a PNG in the app’s private files directory. Cropping also leaves a JPEG in cache, and a photo taken through the system camera can add a temporary PNG. The code that deletes one history record removes the database row, but we found no corresponding deletion of these image files.

That means deleting an entry appears to remove it from the visible history while leaving private image copies in app storage until cached files are reclaimed or the app’s data is cleared or uninstalled. This is a source-code finding, not a result from observing storage on a real phone. For sensitive scans, clear the app’s storage after exporting any text you need.

The manifest also allows Android backup. Whether private app data is copied depends on the device, Android version and backup settings.

Recognition limits

OCR quality depends on the language model, print quality, orientation, lighting and layout. The app shows an accuracy value, but it is not a guarantee that names, amounts or serial numbers are correct. Review important text against the image. Upstream also mentions mathematical-equation detection as not yet tested, so this page does not present it as a dependable feature.

About this file

• Version 6.1, build 61, is 35.0 MB and needs Android 6.0 or later. It includes native libraries for 32- and 64-bit ARM and x86 devices.
• Its SHA-256 is 8fe496fbd91ad57bba697027f86f942d217687f98df63a20d3f3be8ff5aba79c, matching F-Droid’s signed index. F-Droid signs this build.
• The app is Apache-2.0 licensed. The separately downloaded language files come from the Tesseract project.

CPU types in this build: arm64-v8a (64-bit ARM, almost every phone since 2017); armeabi-v7a (32-bit ARM, older and budget phones); x86 (32-bit Intel, emulators); x86_64 (64-bit Intel, emulators and Chromebooks).

Versions currently kept in the F-Droid repository. We host and check only the newest one. Full changelog.