Fig. 1. System framework diagram

Fig. 2. Auxiliary image outputs. (a) Original image; (b) focus image; (c) depth of field image; (d) Mask corresponding to vehicle on left; (e) Mask corresponding to vehicle on right

Fig. 3. Diagram of ResNet-18 network structure

Fig. 4. Filter out Mask that is too small. (a) Original image; (b) (c) Masks before filtering; (d) Mask after filtering

Fig. 5. Filter out overlapping Masks. (a) Original image; (b)‒(e) Masks before filtering; (f) (g) Masks after filtering

Fig. 6. Background blur images. (a) Background blur image after simple fusion; (b) background blur image after edge optimization

Fig. 7. Comparison of results of experiment 1. (a1)(a2) Original image; (b1)(b2) Mask R-CNN; (c1)(c2) YOLACT

Fig. 8. Comparison of results of experiment 2. (a1)(a2) No edge optimization; (b1) (b2) Poisson fusion edge optimization; (c1)(c2) mean filter edge optimization

Fig. 9. Comparison of results of experiment 3. (a1) (a2) Original images; (b1)(b2) small filter cores; (c1)(c2) moderate filter cores; (d1) (d2) large filter cores

Fig. 10. Comparison of results of experiment 3. (a) Depth of field image; (b1)(b2) original images; (c1) without sense of hierarchy (auto focus); (c2) without sense of hierarchy (independent selection); (d1) with sense of hierarchy (auto focus); (d2) with sense of hierarchy (independent selection)

Fig. 11. Comparison of results of experiment 5. (a) Original image; (b) autofocus mode of our method; (c) independent selection mode of our method; (d) Huawei P20; (e) iPhone XR

Fig. 12. Comparison of results of experiment 6. (a) Original image; (b) our method; (c) method proposed by Yang et al.; (d) method proposed by Xiong et al.; (e) method proposed by Purohit et al.; (f) method proposed by Dutta et al.; (g) method proposed by Zheng et al.