A Study on Frame Position Detection of Digitized Comics Images

Recently, comic is read not only through paper media, but also digital devices i.e. personal computer, cell phone, epaper reader and etc. Thus, the display environment is quite different at each device. Comic consists of many frames that have text, character and illustrations. Moreover, the reading order of comic is defined by frames. Therefore, in order to provide user centric presentation functions for various types of digital devices, and facilitate the developments of other high level contents application i.e. auto text extraction, copyright protection system and so on, accurate frame shape extraction is needed. There are many types of document analysis methods which utilized for general documents. For example utilizing histograms, connected components, density of pixels and so on are proposed [1-3]. However, comics have characteristics, which are different from other general document. For instance, unlike regular document structure, comics have free arrangement where some content protrusion from the frame borders is allowable. Some comic frame separation methods have been proposed. Tanaka et al. have proposed a comic segmentation method by iterative separation [4]. We have extended this iterative separation method utilizing consistency of density gradient, and proposed frame corner detection method [5-6]. However, most of results by frame separation method have redundant blanks. Moreover, results of frame corner detection cannot be used alone for determining frame shapes. In this paper, we propose a frame position detection method that utilizes the result of frame separation and frame corner detection for comic images.