Recognition of Layout-Free Characters on Complex Background

Abstract

Recognizing characters in a scene is a challenging and unsolved problem. In this demonstration, we show an effective approach to cope with the problems: recognizing Japanese characters including complex characters such as Kanji (Chinese characters), which may not be aligned on a straight line and may be printed on a complex background. In the demo, our recognition method is applied to image sequences captured with a web camera. The recognition method is based on local features and their alignment. In addition, using a tracking method, recognition results and extracted features are accumulated so as to increase recognition accuracy as time goes on. The demo runs about 1 fps on a standard laptop computer.