A computer vision python program to resolve sudoku taken from a camera
- data : Contains 10 folders, one for each digit (1167 examples)
- examples : Contains images at different stage of the processing
- ressources :
- img : Input image(s) to test
- models : Different Keras CNN models for digit classification
- src : All the python scripts
- extracter.py : The entry point, where all the magic happens
- solver.py : Algorithm(s) to solve the sudoku
- alterateImages.py : Function(s) to alterate images (crop, skew, rotate)
- addBadImages.py : Add 150 alterated images to each digit class
- buildDataset.py : Is used by buildMnist.py to create a dataset in memory
- buildMnist.py : Create, train, test and save a CNN model
- Keras==2.2.4
- numpy==1.15.4
- imutils==0.5.2
- opencv_python==3.4.3.18
- scikit_learn==0.20.1
- Re-train CNN with MNIST data (hand writtten digit) & test perfomance
- Add a web interface
- Use an RCNN to localize digits