Fig. 1. Experimental setup of CGI

Fig. 2. Diagram of proposed algorithm

Fig. 3. Results of edge detection of “rice” using numerical simulation. (a) Original image of “rice”; (b) CGI image under 40000 samples; (c) edge detection image of CGI image under 40000 samples; (d) result of SSGI algorithm under 12000 samples; (e) result of Kirsch_max algorithm under 12000 samples; (f) result of Kirsch_ave algorithm under 12000 samples; (g) edge detection image of original image

Fig. 4. SNRs of different algorithms

Fig. 5. Analysis on anti-noise performances of different algorithms. (a) Results of different algorithms; (b) analysis on anti-noise performances of different algorithms

Fig. 6. Results of edge detection of “lena” using numerical simulation. (a) Original image of “lena”; (b) CGI image under 50000 samples; (c) edge detection map of CGI image under 50000 samples; (d) result of SSGI algorithm under 12000 samples; (e) result of NSCT algorithm under 12000 samples

Fig. 7. Results of edge detection of SSGI and proposed algorithms. (a)-(e)Recovery images of SSGI algorithm; (f) edge detection image of original image obtained by Sobel operator; (g)-(k) recovery images of NSCT based ghost imaging method when threshold TH is 0; (l) edge detection image of original image obtained by NSCT; (m)-(q) recovery images of NSCT based ghost imaging method when threshold TH is 0.17

Fig. 8. MSE of SSCI and different threshold algorithms