These experiments reported here have been performed on 1020training samples and 420 testing samples. And in our experiments,every method is utilized for color correction with 5 times training,while each training result is kept as a color correction model. The experimental result of this paper is shown in Table 1.Based on visual experience, the general criteria for color difference are as follows: when E < 3, the color difference can be sensed; when E < 5, the color correction accuracy can be accepted.In the case of images acquired by real scanners, the noise is naturally present, and the color transformation matrix may amplify this noise. From Table 1, it can be observed that the testing error sproduced by using MEC-BP-Adaboost neural network model are less than by using other models. Compared with the conventional