For the problem of the characteristics of the grey and white granular background image condition on the surface of the plate metals, onedimensional and twodimensional Wellner adaptive threshold algorithm are the first time to be applied to the scene. Based on those two kinds of algorithm, this paper proposes a gaussian weighted adaptive threshold algorithm to solve the problem. Firstly, this algorithm calculates the gaussian weighted distance between pixels in a window to form a weighted distance diagram and then use the ideology of Wellner "centeraround comparison" to calculate binary image directly. Finally, experiments are carried out on images acquired on actual production line. The twodimensional otsu algorithm, uniformity measurement algorithm, onedimensional, twodimensional Wellner adaptive algorithm and the last improved algorithm are compared. Experimental results show that, compared with other algorithms, the algorithm in the end of this paper has obvious advantages in the image segmentation effect.