Fig. 1. Schematic diagram of feedback GI. The picture shows the flow diagram of ELFGI, which is divided into four steps. The arrows show the direction of the steps and data. The red arrow of Step 3 also shows that the illumination patterns are lighted onto the target.

Fig. 2. Experimental setup. The illumination patterns are generated via a laptop (not shown), which controls the emission of a commercial projector. The reflected light from the object is collected with a lens and detected with a CCD camera with the results on all the pixels summed up as a bucket detector.

Fig. 3. Experimental results for two targets shown in (a1) and (c1). (a2)–(a4) and (c2)–(c4) show imaging results via GI using random speckles, with the size of speckles being 16, 4, and 2 pixels, respectively. The number of measurements is 2000. (b1)–(b4) and (d1)–(d4) show results of ELFGI with T=12 obtained at the 4th–7th round, costing 70, 166, 550, 1318 and 43, 139, 523, 1291 frames, respectively.

Fig. 4. Simulation results. (a1) is the target for resolution test, with the width of the narrowest stripes being 1 pixel. (a2) shows results of ELFGI with T=0, (a3) is obtained via GI with random speckles, and (a4) shows results of GI with Hadamard patterns. The number of samplings is 4480, 47,104, and 65,536, respectively. (b1) is a grayscale target of three-level values. (b2)–(b4) are the results of ELFGI with T=0, obtained in the 4th, 6th, and 8th rounds under 256, 1408, and 6016 samplings.

Fig. 5. Simulation results under different noise with different methods. The amount of samplings is 16,384, 65,536, 262,144, and 1,048,576 for traditional GI and 5829, 10,528, 17,984, and 21,056 for ELFGI with T = 0, 76, 152, 304, respectively. As for CSGI, it costs 6000, 8000, 16,000, and 20,000 samplings.