AI_Sudoku

GUI Smart Sudoku Solver that tries to extract a sudoku puzzle from a photo and solve it.

Installation

Download and install Python3 from here
I recommend using virtualenv. Download virtualenv by opening a terminal and typing:
```
pip install virtualenv
```

Create a virtual environment with the name sudokuenv.

Windows

virtualenv sudokuenv
cd sudokuenv/Scripts
activate

Linux:

source sudokuenv/bin/activate

Clone this repository, extract it if you downloaded a .zip or .tar file and cd into the cloned repository.
- For Example:
```
cd A:\AI_Sudoku-master
```
Install the required packages by typing:
```
pip install -r requirements.txt
```

Usage

Before running the application, know that you can set the modeltype variable in Run.py to either "CNN" or "KNN" to choose the Convolutional Neural Network or the K Nearest Neighbours Algorithm for Recognition. By default it is set to "KNN" and I got a way higher accuracy using KNN itself, so I would recommend that you don't change it.

'''Run this file to run the application'''
from MainUI import MainUI
from CNN import CNN
from KNN import KNN
import os
# Change the model type variable value to "CNN" to use the Convolutional Neural Network
# Change the model type variable value to "KNN" to use the K Nearest Neighbours Classifier
modeltype = "KNN"

Type the below command to run the Application. You need to be connected to the Internet and it might take 5-10 minutes to create the knn.sav file so please wait patiently. This delay is only during the first run as once created, the application will use the local file
```
python Run.py
```
The GUI Homepage that opens up as soon as you run the application.
You need to select an image of a Sudoku Puzzle through the GUI Home Page.
Once you press Next, a number of stages of image processing take place which are displayed by the GUI leading up to recognition. Here are snapshots of two of the stages:
For recognition, a CNN or KNN can be used. This option can be toggled as mentioned in the first point. Once recognized, the board is displayed and you can rectify any wrongly recognized entries in the board.
Finally click on reveal solution to display the solution.

Working

Image Preprocessing

Gaussian Blurring Blurring using a Gaussian function. This is to reduce noise and detail.
Adaptive Gaussian Thresholding Adaptive thresholding with a Gaussian Function to account for different illuminations in different parts of the image.
Inverting to make the digits and lines white while making the background black.
Dilation with a plus shaped 3X3 Kernel to fill out any cracks in the board lines and thicken the board lines.
Flood Filling Since the board will probably be the largest blob a.k.a connected component with the largest area, floodfilling from different seed points and finding all connected components followed by finding the largest floodfilled area region will give the board.
The largest blob a.k.a the board is found after the previous step. Let's call this the outerbox
Eroding the grid a bit to undo the effects of the dilation on the outerbox that we did earlier.
Hough Line Transform to find all the lines in the detected outerbox.
Merging related lines. The lines found by the Hough Transform that are close to each other are fused together.
Finding the Extreme lines . We find the border lines by choosing the nearest line from the top with slope almost 0 as the upper edge, the nearest line from the bottom with slope almost 0 as the lower edge, the nearest line from the left with slope almost infinity as the left edge and the nearest line from the right with slope almost infinity as the right edge.
Finding the four intersection points. The four intersection points of these lines are found and plotted along with the lines.
Warping perspective. We find the perspective matrix using the end points, correct the perspective and crop the board out of the original image.
Thresholding and Inverting the grid. The cropped image from the previous step is adaptive thresholded and inverted.
Slicing the grid into 81 slices to get images of each cell of the Sudoku board.
Blackfilling and centering the number. Any white patches other than the number are removed by floodfilling with black from the outer layer points as seeds. Then the approximate bounding box of the number is found and centered in the image.

Recognition

Convolutional Neural Network

Read about CNNs here

Layers A Convolution Layer, a Max Pooling layer flattened into a hidden layer followed by some Dropout Regularization, another hidden layer and finally the output layer. Each of the inner layer uses ReLu while the output layer uses softmax.
Compilation Adam optimizer and sparse categorical cross entropy loss.
Training The model is trained on the MNIST handwritten digits dataset which has around 70,000 28X28 images.
Accuracy Around 98 percent accuracy on the test set.

K Nearest Neighbours

Read about KNN here

K value used is 3.
Training Trained on the MNIST handwritten digits dataset which has around 70,000 28X28 images.
Accuracy Around 97 percent accuracy on the test set.

ToDo

Improve Accuracy.
Resolve Bugs/Issues if any found.
Optimize Code to make it faster.

Contributing

Contributions are welcome 😄

Pull requests

Just a few guidelines:

Write clean code with appropriate comments and add suitable error handling.
Test the application and make sure no bugs/ issues come up.
Open a pull request and I will be happy to acknowledge your contribution after some checking from my side.

Issues

If you find any bugs/issues, raise an issue.

Prashu94/AI_Sudoku