A character recognition apparatus which performs character recognition with increased accuracy on a document image including plural languages. A re-recognition range is set based on the result of recognition using a first recognition unit, and character recognition by a second recognition unit is performed within the set range. In the re-recognition range, if a similarity of the result of re-recognition is higher than that by the first recognition unit, the result of recognition by the first recognition unit is replaced with the result of recognition by the second recognition unit.