Abstract: Simple linear iterative clustering (SLIC) can be applied directly to spherical images in equirectanguler projection (ERP) form. However, the damage of the correlation of the spherical data caused by projections leads to inappropriate superpixels in some areas of the ERP image, which impacts on the performance of the algorithm. To address this issue, resampling for ERP images was first applied to generate spherical image elements which are nearly uniformly distributed on the sphere. Then we rearranged those resampling data to form a novel 2D representation of a spherical image while maintaining the local correlation of spherical image data. Based on such a 2D representation, we integrated the geometrical relations of the spherical data into the SLIC algorithm and finally built a spherical image-based SLIC algorithm. The SLIC algorithm and the proposed algorithm were respectively applied to several groups of ERP images, and the superpixel segmentation results with different clustering numbers generated by the two algorithms were compared. The experiments suggest that the proposed spherical image-based SLIC algorithm outperforms the original algorithm in terms of objective quality, and can generate superpixels without the effect of the variation of the regions on the sphere. The generated superpixels also have closed-contours and present better similarity and consistency on the spherical surface.