Creating a chessboard mask to check the status of the squares on a chessboard

Question

I want to extract the squares from a chessboard and label each one of them according to their coordinates i.e. a1, a2, ... h8. My goal is to create a 'mask' to check the presence of a piece for each square. I am currently writing a program to do this in C# using Emgu CV.

The image used for testing can be found here.

With no experience in computer vision, I have only been following simple code examples guided by basic ideas. My first lead was this tutorial which tells me how to make a simple shape detector. It was not very accurate though as the test image is more complex than the example image.

Inaccurate square detection (with a slightly different image)

There were two problems with the results: 1) not all of the squares were detected as squares; 2) 91 boxes (instead of 64) were created to visualise the detected squares.

To solve problem #1, I used a binary threshold with dilation and median smoothing to further 'simplify' the image and the results were much better but not complete (due to noise):

After applying a binary threshold, dilating, median smoothing and eroding

Visualizing and counting the squares

It appears that the same square(s) is detected multiple times (i.e. problem #2) for reasons unknown to me and I have no idea how to fix this. My initial idea was to give each square a label as soon as they were detected but it doesn't seem practical now.

Questions

1. What method(s) can I use to remove the post-filter noise on squares b3 and g6?
2. What method(s) can I use to properly count and sort the squares in row-by-row order?
3. Assuming that I manage to label each square, how do I then 'save' the region of the square to be reused as a mask for newer frames? Is it computationally expensive to have 64 unique masks?
4. How can I maximize the detection rate of squares?

Answer 1

Regarding my comment in the original question, I implemented something that might serve as an example of fixed grid construction:

#define SQUARE_SIZE 131
#define OFFSET_X 75
#define OFFSET_Y 75
#define NUM_SQUARES_X 8
#define NUM_SQUARES_Y 8

using namespace std;
using namespace cv;

int main(){

    // Directory settings
    const string pathToData = "../data/";

    vector<string> filenames = std::vector<string>();

    // Load filenames
    glob(pathToData, filenames, false);

    if(filenames.size() <= 0){
        cout << "no images were found..." << endl;
        return -1;
    }


    Mat chessboard = imread(pathToData+"chessboard.jpg", IMREAD_UNCHANGED);

    for (int i=0;i<NUM_SQUARES_X;i++){
        for (int j=0;j<NUM_SQUARES_Y;j++){

            int startX = OFFSET_X + SQUARE_SIZE*i;
            int startY = OFFSET_Y + SQUARE_SIZE*j;
            int endX = startX + SQUARE_SIZE;
            int endY = startY + SQUARE_SIZE;

            rectangle(chessboard, Point(startX,startY), Point(endX,endY), Scalar(255,0,0), 3, 8, 0);
        }
    }

    namedWindow("chessboard", WINDOW_NORMAL);
    imshow("chessboard", chessboard);  

    cv::waitKey(0); 

    return 0;
}

The output of the above code can be seen in the image below:

Of course you'll have to fine tune the parameters (offset and square dimensions), since I didn't spend much time on it.