The invention discloses a super-pixel segmentation method for a vision degradation image. k square superpixels with the same size are pre-segmented on the RGB image and the corresponding depth image respectively; taking the centers of the K pre-segmented superpixels as initial seed points; dividing an initial rectangular clustering region by taking the initial seed point as a center; distributinga color information weight Qc and a depth information weight Qd for the clustering area by using the color edge strength and the depth edge strength of the clustering area; and calculating a superpixel clustering distance Dp according to the color distance Dc, the depth distance Dd, the spatial distance Ds, the color information weight Qc and the depth information weight Qd, then performing re-clustering and repeated iterative convergence, and generating superpixels according to the converged clustering area. According to the method provided by the invention, the edge fitting rate of the superpixels and the object under the condition of visual degradation of the image is obviously improved.