Extraction of relevant information from document images using measures of visual attention

This paper describes an approach to attention based layout segmentation using general principles of the human visual perception to achieve this goal. The text is considered as texture in different resolution levels. A new measure of attractiveness is introduced. The segmentation is generic and not limited to specific document classes and models. The resulting regions of interest may be used for further interpretation. The overall speed of browsing and searching large volumes of scanned documents can be increased considerably.