Abstract: This paper proposes the 3 thresholds edge detection algorithm based on multi-scale integration in the wavelet domain, the advantage of this method is the excellent anti-noise performance. It gives the concept of gradient direction characteristics utilizing the differences between the edge point and noise point along the gradient direction. First of all, the wavelet transform for the image on the different scale hasbeendone to get gradient information of each pixel. Then it detects the edge point through the non-maximum inhibition method of dual-threshold and the gradient direction characteristics. At the last, the edge extracting results from the integration of the third threshold and the detection results in each scale are obtained. Compared with the classical Canny and Mallat operators, experiments denominate this algorithm not only has the good capacity of edge location, but also can extract the satisfactory results for the image involved with strong noise.