Host: Japan Society for Fuzzy Theory and Intelligent Informatics (SOFT)
Mobile devices have a function of character recognition in recent years. Due to the function to acquire text information, it is possible to use the mobile devices as the way to use geographical information system, image search and classification engines, etc. However, it is difficult to recognize characters in natural scenes, because there is too much information in these scenes. In the earlier work, a binarization-based method to extract a signboard from those scenes has been reported. There is a disadvantage that this technique depends on colors of text and background. In this study, we present a method of signboard extraction that doesn't depend on the color of text and background. The proposed method employs a stereo image to extract a signboard, and performs a three-dimensional analysis of the scene. The effectiveness of the proposed method was shown by applying to some natural scenes.
