Fig. 1. SPIRAL algorithm using the l1-norm, TV, and l1-TV. FGP, fast gradient projection.

Fig. 2. MSE convergence profiles of the l1-norm, TV, and l1-TV.

Fig. 3. RMSEs of the reconstructed images using the l1-norm (green square), TV (blue diamond), RPD-Ti (purple dot), and l1-TV (red triangle).

Fig. 4. Comparison of three groups of reconstructed images using different penalties. From left to right are ground truth (GT), l1-norm, TV, RDP-Ti, and l1-TV. The RMSEs between the reconstructed images and their corresponding ground truth are listed below, respectively.

Fig. 5. Comparison of reconstructed depth information using different penalties. From left to right are ground truth (GT), l1-norm, TV, RDP-Ti, and l1-TV. The RMSEs between the reconstructions and the ground truth are listed below, respectively.

Fig. 6. Schematic of experimental system. A repetitive pulsed laser is modulated by a digital micromirror device (DMD) and raster-scans the object pixel by pixel. The reflected photons from the object arrive at a photomultiplier tube (PMT) and trigger a photon-arrival event according to the photon detection probability. The first-photon data are then recorded by a time-correlated single-photon counting (TCSPC) module and transferred to a computer for image reconstruction.

Fig. 7. Comparison of experimental results yielded by different approaches. From left to right are photon counting (PC), l1-norm, TV, RDP-Ti, and l1-TV. The RMSEs between the reconstructed images and the photon counting image are listed below, respectively.

Fig. 8. Comparison of experimentally reconstructed depth information using different penalties. From left to right are ground truth (GT), l1-norm, TV, RDP-Ti, and l1-TV. The RMSEs between the reconstructions and the ground truth are listed below, respectively.