BACKGROUND
Character and document processing have a long history in image processing and computer vision field. Typical examples are OCR (Optical Character Recognition), document retrieval, or digital library. Some of these applications show more or less satisfactory results or performances in real-life. A lot of research has still been being made on these topics However, they often assume that they have clean document images or they could capture a region where the required information is located. For instance, the experimental dataset of [3], [4], and [5] have already text or paper images of a document. The test images in [6], [7] and [8] has also simple background or document images.Song et al. proposed a method of detecting text regions 19, but it also deals with well-captured document images or images with text regions without other background objects. Lampert et al. proposed a similar system as the proposed method in [10], but cameras are fixed over a desk. The background of captured images could be modelled statistically in this case. It is easier to detect a paper region with this statistical model compared with using a single captured image. In the proposed method, a digital camera need not to be fixed in somewhere. Also, it does not require any statistical modelling process, which means a single captured image is enough to capture a document from the image.