Abstract: For defective rusting on nuclear containment image, with lots of noise and interference similar in color, in this paper a rusting recognition algorithm of contour directed dilation is proposed. Using the characteristic yellowish orange color, extract basic rusting information through RGB multi-channel threshold segmentation. On the basis of rusting's annulus shape feature, the rusting by multi-direction according to contour's extension direction is divided. Contour fragment is dilated by the self-design eight-direction kernels in order to complement annulus rusting. The real rusting is recognized by roundness calculated according to contour moments. Experiments show that the algorithm can undo amounts of noise and interference similar in color so as to recognize and position the real rusting on the image.

Key words: rusting recognition, threshold segmentation, directed dilation, roundness recognition, rusting position