By grouping pixels with visual coherence, superpixel algorithms provide an alternative representation of regular pixel grid for precise and efficient image segmentation. In this paper, a multi-stage model is used for sea ice segmentation from the highresolution optical imagery, including the pre-processing to enhance the image contrast and suppress the noise, superpixel generation and classification, and post-processing to refine the segmented results. Four superpixel algorithms are evaluated within the framework, i.e. SLIC, BASS, TS-SLIC and WP, where the high-resolution imagery of the Chukchi sea is used for validation. Overall, the model yields a segmentation accuracy of 98.19% on average and adhere the ice edges well. We also present quantitative evaluation in terms of the segmentation quality and floe size distribution, and visual comparison with several selected regions of interest. It is found that TS-SLIC performs the best within the group.