Abstract: A method to improve the competence of fuzzy morphological associative memories against random noise is presented, through the decomposition of gray-scale images, combining the method of fuzzy morphological associative memories. With the method, the auto-association and recognition of gray-scale images with random noise is settled. Meanwhile, this method has been extended to the settlement of the case of color image. Accordingly, a better way to resume gray-scale images and color images with noise is offered. Our experimental results here demonstrate the effectiveness of this approach.