jackploeg / android-sudoku-solver-ocr Goto Github PK

• Introduction
• How to use
• How it works
• Build
• Libraries used

This project is an android app that solves sudoku, hence the name, recognized from images using Optical Character Recognition or entered manually.

I've built this project the first time in early 2022 ( v1 branch ) and this is a refresh, an update to the original project.

Download or build the app yourself. Use it like a camera app, press the shutter button when the sudoku it's detected and it will be solved. You can also edit it or input it manually if you want.

It works on both puzzles on paper or on screen, the only requirement is there is enough light and that it's smooth.

To identify the sudoku board the input image is processed as follows with OpenCV:

• Using the Canny algorithm to detect all contours in the image.
• The biggest contour corresponds to the whole sudoku puzzle.
• Use the found contour to extract and warp the sudoku grid .

The numbers are detected using MLKit Text Recognition.

For solving the puzzle I'm using a fairly simple, brute-force algorithm that relies on backtracking to generate the valid solution.

It goes through the whole 2D array and for each number that needs to be found it tries all possibilities and continues with the following numbers.

For each cell there are 9 possible numbers which means the time complexity of this algorithm is O(9ⁿ).

private fun solve(index: Int = 0): Boolean {
    for (i in index until 81) {
        val r = i / 9
        val c = i % 9
        if (sudokuBoard[r][c].number == 0) {
            for (n in 1..9) {
                if (isValidSolution(r, c, n)) {
                    sudokuBoard[r][c].number = n
                    if (solve(i + 1)) {
                        return true
                    }
                    sudokuBoard[r][c].number = 0
                }
            }
            return false
        }
    }
    return true
}

Solve algorithm demo

• Download or clone the repository and open it in Android Studio.
• Download the android version of OpenCV. I used the latest version, 4.8.0.
• Import the OpenCV module in Android Studio.
  • Click File > New > Import Module, select the OpenCV-android-sdk/sdk directory and change the module name to opencv.
  • Goto File > Project Structure > Dependencies and on the app module add opencv as a dependency.
  • In opencv build.gradle add android { namespace 'org.opencv' }, change compileSdkVersion and targetSdkVersion to 33 and finally, add

    kotlinOptions {
      jvmTarget = '1.8'
    }
    buildFeatures {
      buildConfig true
      aidl true
    }
    

  • Further we need to modify the following classes in the OpenCV module:
    • CameraBridgeViewBase: add canvas.rotate(90f, canvas.getWidth() / 2, canvas.getHeight() / 2); in function deliverAndDrawFrame, right before drawing the bitmap.
    • JavaCameraView add:

      public void turnOnFlashlight(){
          Camera.Parameters params = mCamera.getParameters();
          params.setFlashMode(Camera.Parameters.FLASH_MODE_TORCH);
          mCamera.setParameters(params);
      }
          
      public void turnOffFlashLight(){
          Camera.Parameters params = mCamera.getParameters();
          params.setFlashMode(Camera.Parameters.FLASH_MODE_OFF);
          mCamera.setParameters(params);
      }
      

      In function iniliazeCamera change

      if ((getLayoutParams().width == LayoutParams.MATCH_PARENT) && (getLayoutParams().height == LayoutParams.MATCH_PARENT))
          mScale = Math.min(((float)height)/mFrameHeight, ((float)width)/mFrameWidth);
      else
          mScale = 0;
      

      to

      mScale = Math.max(((float)height)/mFrameHeight, ((float)width)/mFrameWidth);
      

• Done, you can now build the project

• OpenCV for image processing.
• ML Kit Text Recognition V2 for getting the numbers out of the image.