Many factors affect the accuracy of detection, especially when it is a digital picture (picture of a computer screen, monitor etc). This is because of the digital noise that stems from interference when the camera is taking a snapshot.
- Before capturing the picture, adjust as much as possible to reduce noise seen in the camera preview - This is usually achieved when the board is as big as possible in the camera preview
- Screenshot the board and import via the gallery feature. This removes digital noise interference completely.
- Stabilise camera and allow it to auto-focus before snapping
- Adjust such that sudoku board fills most of the area in the camera preview
There is a delete icon at the right of the appbar (top bar) in the details page. Alternatively, you may swipe left on the history entry in the history page.
Depends on the number of sudoku boards available for the model to be trained effectively. I am looking at once a month. The current models are very accurate in detection already.
Please view the contributing section in README.md and CONTRIBUTING.md.
You may leave your donation by navigating to the donation page in the app Settings -> About -> Donate. THank you in advance!
Snap Solve Sudoku V1 utilises Tesseract 3 for digit recognition while V2 uses a custom convolutional neural network machine learning model via Tensorflow and Keras. This enables improvement for recognition to be made easily and on the fly. You can update the model directly from the app whenever there is one available!
Completely rewritten in Kotlin, Snap Solve Sudoku V2 utilises coroutines as much as it is feasible to speed up processes in image processing. Although V1 utilises multithreading as well, coroutines make it much easier and less cumbersome.
UI and UX wise, V2 is more appealing and more modernised.