Abstract: Nowadays the fusion rules of the images of strip surface defects are one-sided, since they are based on the single point coefficient and the features of regional coefficient, thus making the images easy to be distorted. To solve the problem, the new fusion rules are presented where the high-frequency sub-bands are fused together after wavelet decomposition, with the dissimilarity between the local variances of the two images to be fused taken fully into account, then the thresholds are set to select the appropriate fusion rules. Experimental results showed that the method proposed can keep up the image features in detail and is highly applicable to the image fusion of surface defects of cold-rolled strip.