A method is described for extracting text regions from scenery images captured by digital camera. Most methods proposed for extracting such regions use a specific feature, such as an edge feature, so they must use many conditions for extraction. It is also difficult to assign the conditions. A method described can use a few conditions because it uses two kinds of features. Closed edges which reflects local aspects of an image in the image domain; are used for extraction. Color clustering, which reflects the global aspect, is used for segment ation. In each feature image, candidate character regions are selected based on the relative geometric data among the regions. Regions to the selected one are extracted as character regions and used to reconstruct the text regions. Character recognition experiments using commercial software are shown.
