Clarity Through Complexity
This solver reads a Sudoku puzzle through your camera. Your video is processed only to detect and solve the puzzle — nothing is stored.
Hold a printed Sudoku puzzle flat and steady in the frame.
Need a puzzle? Print one here →
Your webcam streams to the browser. A few frames per second are sampled and sent to the solver — nothing is stored.
OpenCV locates the puzzle's grid by its largest square contour, then warps the perspective into a flat, top-down view.
The grid is split into 81 cells. A convolutional neural network reads each digit and reports a confidence score.
Once recognition confidence clears 99%, a backtracking algorithm solves the board and the answer is projected back onto your puzzle.