EXTRACTION OF SUPERIMPOSITION WHEN THE COLOR OF CHARACTERS ARE SIMILAR TO THE BACKGROUND

摘要

Superimposition in videos is one of the most important information to make indexes of image contents automatically. In the case of license plates or signboards, it is easy to distinguish characters from the background by using the brightness. The reason is that they usually have two different colors assigned to characters and the background. On the other hand, in the case of superimposition, it is difficult to separate them by using only the brightness because it has not a monotone background color like signboards. In this paper, we propose a new method to extract superimposition from videos based on the information of edge and labeling images. Applying the method to 19 kinds of super imposition of a real video, we extracted almost every characters.