A method and device are described for extracting metadata from an image of pixels, such as a title or author of a document. At least part of the image is shown on a display for a user. A pointing control element in a user interface, such as a mouse or a touch screen, is operated by a user to generate a selection command. The selection command includes a selection point in a metadata element in the image. A region of foreground pixels is determined, the region containing pixels that are connected to the selection point. An extraction area is constructed around the region. Finally metadata is extracted by processing pixels in the extraction area.