Affine invariant information embedment for accurate camera-based character recognition

Recognizing characters in a scene image taken by a digital camera has been studied for decades. However, it is still a challenging problem to achieve high accuracy. In this paper, we propose a method of embedding information in a character pattern so that the class of the character can be identified. The information should be robust against geometric distortions since an image taken by a digital camera is usually geometrically distorted. In the proposed method, a character pattern is designed in two colors so that the information is embedded as the area ratio of regions of two colors. Since the area ratio is affine invariant, it is expected that the area ratio is correctly extracted even if a character image is affine-transformed. We generate character patterns with the embedded information and discuss the effectiveness of the proposed method.