| Summary: | 碩士 === 立德管理學院 === 應用資訊研究所 === 91 === Textual data generally carry useful and important information in human’s daily life, such as TV programs, newspaper, magazines, etc. In this thesis, a new method that is the combination of the morphological technique and the smooth gray level detection is proposed for locating text in images from different sources, including magazines, photographs, WWW images, movie posters, and videos. First, the RGB components of the input color image are combined to give an intensity grayscale image, and then the morphological gradient is used to detect edges in the grayscale image. The connected component algorithm is used to extract these connected components, and then some heuristic criteria are used to combine the neighboring components. The smooth gray level detection and new RLSA are proposed for identifying candidates for the text regions. Furthermore, the non-text regions are eliminated according to the text stroke thickness. Finally, Restore lose characters (RLC) is applied to restore lost characters according to the neighboring consensus. From the experiments, the proposed method on locating text is insensitive to the font-style, font-color, font-size, orientation, language, and background complexity.
|
|---|
