Sudoku

Sudoku solver using Computer Vision and Deep Learning.

Computer Vision

Using OpenCv to detect the board from a video stream.
Getting a bird eye view of the board and extacting the cells.

Deep Learning

first we create synthetic digits images

Creating 28x28 images of digit using pillow, and augmentating the data using OpenCv to create different representation of digits.
The augmentation includes different fonts, rotation and Morphological Transformations
such as erode, opening, closing,and sharpening.

Expanding digit database using generative adversarial network

Taking the synthetic digits images that we created and feeding them into a gan to create more variance in our digit database.

Training a Deep Learning Network on our newly created database

Using TensorFlow and a simple neural Network to create a digit classifier. I didnt used mnist on purpose in order to expirment with GAN's.
At the end we got a comparable preforming model to the one trained on mnist.