An image processing apparatus detects boundaries from an image; extracts straight-line segments from the boundaries; detects a region where differences between pixel values of near pixels located across the segments are larger than, or at least a predetermined first value; classifies the segments in the region into four sides of quadrangles; detects colors or densities in outer areas of the four sides; selects combinations that differences between the colors or the densities corresponding to the segments in the combinations are no more than, or smaller than a predetermined second value, from the combinations of the four sides that possibly form the quadrangles; detects coordinates of four vertexes obtained when the segments in the combinations are extended, and a combination that an area of a quadrangle formed by the corresponding four coordinates satisfies a predetermined condition; and corrects the quadrangle formed by the combination to a rectangle.