Fig. 1. Partition results for different k values. (a) Original image; (b) k=50; (c) k=100; (d) k=200

Fig. 2. Extraction of edge features. (a) Original images; (b) Canny operator; (c) Sobel operator; (d) Ref. [14]; (e) proposed algorithm

Fig. 3. Extraction of salient color features. (a) Original images; (b) blurred histogram; (c) global color feature; (d) salient region; (e) results by proposed algorithm

Fig. 4. Structural diagram of network synthesis

Fig. 5. Feature detection based on loop network structure. (a) Original images; (b) salient regions without loop module; (c) salient regions with loop module; (d) ground truth figures

Fig. 6. Results of proposed algorithm and other algorithms. (a) Original images; (b) SUN algorithm; (c) SEG algorithm; (d) RC algorithm; (e) CA algorithm; (f) MZ algorithm; (g) AC algorithm; (h) LC algorithm; (i) FT algorithm; (j) DSR algorithm; (k) HC algorithm; (l) CSD algorithm; (m) GU algorithm; (n) proposed algorithm; (o) ground truth figures

Table 1. Comparison of quantitative results

Table 2. Average computation time of different methods on 10 images