This could just be a case of needing to optimize and train their recognizer to deal with a much larger set of possible characters, or it could require implementing kanji-specific OCR techniques like attempting to decompose the characters into their constituent strokes, and recognize based on classification of those strokes (orientation, position, direction).