Today there are lots of documents in which text characters are printed on colored and/or complex backgrounds. We previously proposed a character pattern extraction method by which character paterns can be extracted from grayscale document images with comlex background. The method has unique, advantageous properties; it is capable of extracting very small characters and is tolerant of shadings of images. However, the method did not work well for some color documents since it lacks the ability of discriminating color difference. This paper proposes an enhanced version of the method, which utilizes the local color segmentation and the region growing. The experimental results have shown that the new method yields much better results for color magazine covers without sacrificing the performance of extracting small character patterns. The method is tolerant of shadings of images as well.