A Contrast Pre-adjusted Defect Detection of Strip Steel Surface by Total Variation-based Image Decomposition

Automatic vision-based defect detection on the strip steel surface is a challenging task due to miscellaneous patterns of defects, low contrast of the image. Traditional methods mainly distinguish the defects from the background by analyzing textural information. In this paper, a novel defect detection method by total variation-based image decomposition is developed to extract abnormal region from repetitive texture. Specially, a contrast-adjusted step is designed for the image decomposition, which can compress the dynamic range of the background brightness and enhance the contrast effectively. Furthermore, the histogram of the enhanced image shows multi-peak characteristics, which greatly contributes to the subsequent image decomposition. Next, the total variation-based image decomposition is developed for the contrast-adjusted image to extract structural map. Finally, the abnormal regions can be located only by an adaptive threshold in the structural map. The proposed method can be extended to detect the defects on other regularly textured surface, even under the low signal-to-noise ratio condition. The experiments show that the performance of the proposed algorithm is better than the state-of-art algorithms.