AUTOMATIC EXTRACTION OF GO GAME POSITIONS FROM IMAGES: A MULTI-STRATEGICAL APPROACH TO CONSTRAINED MULTI-OBJECT RECOGNITION

摘要

Here, we present a constrained object recognition task that has been robustly solved largely with simple machine learning methods, using a small corpus of about 100 images taken under a variety of lighting conditions. The task was to analyze images from a hand-held mobile, phone camera showing an endgame, position for the Japanese board game Go. The presented system would already be, sufficient to reconstruct the full Go game record from a video record of the game and thus is complementary to Seewald (2003), which focuses on solving the. same task using different sensors. The presented system is robust to a variety of lighting conditions, works with cheap low-quality cameras, and is resistant to changes in board or camera position without the need for any manual calibration.