An image extraction device for extracting an image showing a document from an image of the document shot by an imaging device includes a document table on which a document is placed, a side detection unit, an opposing side estimator, and an image extraction unit. The side detection unit detects a side of the document on the document table based on the shot image of the document. The opposing side estimator estimates a position of a side opposite the side detected by the side detection unit. The image extraction unit segments the image of the document from the shot image of the document based on the side detected by the side detection unit and the opposite side estimated by the opposing side estimator.