An Offline Mahjong Support System Based on Augmented Reality with Context-aware Image Recognition

Abstract

The diffusion of the Augmented Reality framework has facilitated the implementation of a support system for a real world task. This paper introduces a system that supports the Mahjong game for beginners. However, it is difficult to calculate the score for beginners in the real Mahjong game. We develop an offline system to get the score by recognizing the Mahjong tiles. This is a difficult task because Mahjong tiles have classes and attributes that are the context of their positions. The system needs context-aware image recognition. The system detected tiles using OpenCV and Convolutional Neural Network to classify them and self-localization indirectly found in attributes. The experimental results show that the system can be used effectively.