A method and an apparatus for searching character image in an image whether the characters are incorporated into an image or separately picked up from a flip card, which are difficult to be detected by the prior art. The invention enables the character strings thereby detected to be searched in a common framework without relying on the language or the character font. An image is chosen, a character region is detected from the frame of the chosen image on the basis of its shape, and a visual feature of the character area is extracted. On the other hand, a character string to be searched for which has been entered by a character input means is drawn as an image, and a visual feature is extracted from the character area image. The visual feature of the obtained character area and the visual feature of the character area image are matched to determine a level of similarity, and the character region containing a character string the level of similarity of which has been determined is outputted.