Sudoku Downloader and Recognizer
Author: Pedro Evaristo González Sánchez
Main Activity(Portrait)
Recognizement
Custom View
Gaming
Backtracking
Board Downloader
Custom view
- Paint and Canvas to draw, background, lines and numbers
- Adaptable ( OnSizeChanged )
- Events Self-management( onTouchEvent, onKeypressed )
- Interaction with Main Activity
Custom View – Main Activity
Main Activity instances Custom View
Main activity set sensorial’s methods of the Custom View (Focusable…)
Whether a Button number is pressed, Main Activity recieves the actual cell position iluminated
Main Activity updates the Sudoku Board and order the custom view to be painted onDraw() and invalidate() )
Class Sudoku (I) Matrix [9][9] of Integers
Constructor overload ( String ) to implement the communication between activities
The class return a String which contains the sudoku board. This is very helpful for several situations. For innstance ( Bundle and Activities communication)
Neighbor checking ( row, column and block )
Sudoku Class II ( Backtracking )
- If the actual Cell is (9,9) => Ends
- If you find a number, go to the next cell ( Original Sudoku )
- For each value 1..9
- If neighbor checking is positive- Sudoku(i+1,j+1)
- If you don’t find a candidate, restore and go back one cell
Download Activity (I)- A way to update the app game
- It exist a Sudoku board holded in a web server
- “Good formed” Board within a *.txt file.
- The goal is to rescue this board and take it to the Game
Download Activity (II)
3. Sudoku Recognizer
3.0 Android previous concepts
3.1 Image Capture
3.2 Image processing (OpenCV)
3.3 Optical Chracter Recognition (OCR) Tesseract
3.0 Android previous concepts
SDK and OpenCV Library (OpenCV)
NDK (Native Developing Kit) ( Tesseract ) Build libraries and reference them in the application.
3.1 Image Capture
- Calling the Intent in charge of capturing images of Android
- Capture and save the image in a temporal file
- Pre-processing => Finding the balance between good qualitiy and computability
3.2 Image Processing (I) First controversy, the image is in perspective => the board is
not a quad, is a trapezoid
The Main concern is a processing in two levels (RGB and GRAY)
Border’s detection
Next, searching the biggest contour within the image. (ROI)
Image Processing (II)
Image Processing (III) Hough algorithm and transform to find the boards lines Two fundamental concepts in line processing:
Every single line will be delete painting them of white or black in each case (RGB and GRAY).
Check if the start point or end point of line is inside some of the four rect corners of the image
Image Processing (IV) Getting the matrix that defines the perspective transformation
needed to extrapolate the corners of the board to the corners of the image
Apply a WarpPerspective Transformation with the matriz obtained.
Image Processing (V)
At this point, the board is divided in equal cells to process each one of them
Procesamiento de la Imagen (VI)
Single cell processing in paralell (RGB and Gray) bearing in mind Tesseracts requeriments.
Equalizing Histogram
Threshold Smooth + Dilate
findContours()
Tesseract ( OCR )
¡Don’t wait for a miracle!. You need an strong image processing
Delimiting dictionary to numbers (1..9)
For each number it is necessary releasing the memory allocated by the image and by the tesseracts object. We are on a phone and we’ve got less memory resources.
Tesseract 6
That’s all. Thank you!!