In the weld seam-tracking system based on machine vision, the welding seam images acquired in the welding process are mixed with strong noises because of the effect of welding conditions. This makes it difficult to get precise seam situation. In this paper, a new edge extraction algorithm based on the fractal theory was introduced to seek welding edge's position in the seam images with noises. In this algorithm, seam images are analyzed depending on the fractional Brownian field model, and the fractal dimension of some pixels and their neighborhood in seam images are calculated by selecting the scaleless range properly, thus distinguishing the seam edge area. The proposed algorithm was then applied to process the images with noises in natural light and arc light (mixed gas shielded welding). The result shows that this algorithm can be used to Correctly extract the position information of welding seam even with strong interference, with the central position error of welding seam less than 0.3mm.