Aiming at laser heating the specimen locally to detect cracks, and the signal-to-noise ratios fell when surface cracks in material with low thermal conductivity were imaged, a detection algorithm of surface microcrack which compared the adjacent thermal signals in cylinder ferrite components was proposed. The thermal signal at each position on moving cylinder was reconstructed based on geometric model, then the Euclidean distance along the laser scanning direction was used as a feature to image. The relative algorithm parameters about cutting area of thermal signal and the position of reference signal were confirmed by simulation analysis. The specimen were test, which had 5-35 μm wide crack on 6 cylinder ferrite surface. The results show that this approach can enhance the signal-to-noise ratio by 1 to 2 times compared with conventional approach at the line-laser scanning speed of 2.66 mm/s, and it can image the shape of cracks evidently.