Fig. 1. Experimental setup of the GISC camera. A conventional imaging system, which consists of the filter, lens 1 with focus length f1, lens 2 with focus length f2, and iris with diameter D, projects the object into its image plane. A spatial random phase modulator (SRPM) is set before detector 2 to modulate the image of the object into a speckle image, which is recorded on detector 2. For comparison, a ground-truth image of the object is recorded on detector 1 through the conventional imaging system with a large aperture diameter of the iris by another optical path via a beam splitter (BS).

Fig. 2. Mutual coherence comparison and recovery performance comparison among DGI, PreGI, TwIST, and PreTwIST algorithms at different sampling rates. (a) Mutual coherence as a function of sampling rate; (b) normalized mutual correlation function of original matrix Φ and preconditioned matrix PΦ; (c) recovery results of DGI, PreGI, TwIST, and PreTwIST algorithms at the sampling rate 0.4 under the detection signal-to-noise ratio (SNR) 25 dB; (d), (e) recovery comparison for resolution target and cell image with a size of 133×133 pixels through the PSNR between DGI, PreGI, TwIST, and PreTwIST at different sampling rates.

Fig. 3. Simulated recovery results by DGI, PreGI, TwIST, and PreTwIST algorithms under different detection SNRs. (a) Normalized mutual correlation function of preconditioned matrix PΦ under different detection SNRs; (b) reconstructed results of DGI, PreGI, TwIST, and PreTwIST algorithms at the sampling rate 0.4 under the detection SNR 40 dB; (c), (d) recovery comparison for the resolution target and cell image with a size of 133×133 pixels through the PSNR between DGI, PreGI, TwIST, and PreTwIST under different detection SNRs.

Fig. 4. Experimental results of DGI, PreGI, TwIST, and PreTwIST algorithms for five different objects with a size of 133×133 pixels, where the (a)–(c) represent the transmission objects and (d), (e) denote the reflective objects. The sampling rate for objects (a), (b) is 0.23, and for objects (c)–(e) is 0.45. The detection SNRs for objects (a)–(e) are, respectively, 24.8 dB, 24.7 dB, 24.8 dB, 20.0 dB, and 23.2 dB.

Fig. 5. Experimental comparison of the PNSR results of the resolution target with a size of 133×133 pixels by using DGI, PreGI, TwIST, and PreTwIST algorithms under (a) different sampling rates and (b) different detection SNRs.

Fig. 6. Comparison of the normalized mutual correlation function of the GISC camera and the theoretical result. The theoretical result is the Rayleigh criterion with incident wavelength 532 nm.

Fig. 7. Experimental results of resolution target with a size of 120×120 pixels for super-resolution imaging. (a) Ground-truth image of the object is recorded by a conventional imaging system with a large aperture diameter of 25 mm; (b) diffraction-limited image of the object via the conventional imaging system with a small aperture diameter of 10 mm; recovery results by (c) DGI, (d) PreGI, (e) TwIST, and (f) PreTwIST; (g) comparison of resolution, intensity profiles extracted from the cross-section black lines in subfigures (b)–(f).

Fig. 8. Experimental results of a butterfly target with a size of 120×120 pixels for super-resolution imaging. (a) Ground-truth image of the object is recorded by a conventional imaging system with a large aperture diameter of 25 mm; (b) diffraction-limited image of the object via the conventional imaging system with a small aperture diameter of 10 mm; recovery results by (c) DGI, (d) PreGI, (e) TwIST, and (f) PreTwIST.