A user specifies a recognition-target area including a character from a target image including the character in the background image. When pixels in the recognition-target area have a color that is close to the color of the pixels on the outline of the recognition-target area, the color of the pixels in the recognition-target area is converted into a predetermined color, that is, the pixels in the recognition-target area are turned transparent, for example. Subsequently, a character-recognition process is performed on the recognition-target area in which the color of the pixels has been converted. According to the device and the method of the present invention, noise and a background image are effectively eliminated.