Abstract: This paper presents an image processing method which can extract the edge of welding seam reliably.Histogram equalization is performed on the gradient image of the original welding image to compress the gray distribution of pixels which represent the edge information of the image.Then the global binary threshold is determined.A small window is sliding on the histogram equalized gradient image with a step equal to half-width of thewindow to determine the local binary threshold in the window area.The adaptive binary threshold in the window area is obtained by combining the global and the local binary threshold.The image is binarized according to the adaptive binary threshold.Moreover, an iterative method is designed to remove noises in the binary image.The testing results of practical images demonstrate the validity of the method.