Fig. 1. Illustration of projection format for panoramic image

Fig. 2. Overall framework of proposed method

Fig. 3. Results of superpixel segmentation of SLIC for global images. (a) Original images; (b) results of superpixel segmentation for K=1200; (c) results of superpixel segmentation for K=600; (d) results of superpixel segmentation for K=300

Fig. 4. Results of superpixel segmentation of local images. (a) Original images; (b) six projection surfaces of CMP; (c) results of superpixel segmentation for K=600

Fig. 5. Original images and saliency maps achieved based on color similarity. (a) Original images; (b) saliency maps achieved based on color similarity

Fig. 6. Original images and saliency maps achieved based on local contrast. (a) Original images; (b) saliency maps achieved based on local contrast

Fig. 7. Procedure of proposed saliency framework

Fig. 8. Illustration of fused saliency map of ERP and CMP maps

Fig. 9. 3D saliency maps. (a) Original images; (b) disparity maps; (c) stereoscopic saliency maps; (d) final fused saliency maps

Fig. 10. Comparison of performance of different saliency detection models

Fig. 11. Precision-recall curves of different saliency models

Fig. 12. Precision, recall, and F-measure of different saliency models