Abstract: Two CCD camera sensors were applied to the same object to acquire different defect images on strip steel surface in several ways, then the images were implemented to those defects through wavelet transformation. The fusion algorithm and rules based on the abstract features from the real surface defects are discussed with the quality of defect images after fusion evaluated. Experimental results showed obviously the superiority of the approach mentioned above because it can acquire more accurate defect image data as much as possible, thus solving the problem of missing the defects due to the acquisition by single CCD camera and providing more useful and reliable data for further process of image recognition and classification.