Image segmentation is one of the difficult problems in computer vision research. Recently spectral clustering has a wide application in pattern recognition and image segmentation. Compared with traditional clustering methods, it can cluster samples in any form feature space and has a global optimal solution. Originating from the equivalence between the spectral clustering and weighted kernel K-means, the authors proposed a spectral clustering algorithm with spatial constraints based on the spatially coherent property of images, also named continuous property. The spatially coherent property means that pixels in the neighbor region should share the same label assignment with the centre one with a high probability. The algorithm adds a term of spatial constraints to the objective function of weighted kernel K-means and makes the minimization of the objective function be equivalent to the spectral clustering through approximation. Experimental results show that our proposed algorithm outperforms the traditional spectral clustering in image segmentation.