Presentation Information
[8a-A23-6]Exposure-Time Reduction in Wavelength-Division-Multiplexed Ghost Imaging Using Averaged Per-Wavelength Reconstructions and Deep-Constraint Denoising
〇Noriki Komori1, Shin Motooka1, Satoshi Sunada1 (1.Kanazawa Univ.)
Keywords:
ghost imaging,high-speed imaging,deep learning
We have demonstrated high-speed single-detector ghost imaging using 25-GHz multimode-fiber speckle illumination multiplexed across five wavelengths (WDM ghost imaging). High-quality reconstruction has relied on a differential scheme that subtracts normal- and inverted-target signals to suppress background and common-mode noise, but this doubles the acquisition time. To avoid this, we combine one-sided acquisition—using only the normal or only the inverted target—with per-wavelength reconstruction: a pseudo-inverse image per wavelength is averaged into an initial estimate, then refined by Ghost Imaging using Deep neural-network Constraint (GIDC) under self-supervised data consistency. For the numeral 0, the differential gave SSIM 0.843, while inverted-only reached 0.782 at half the time (only 0.061 lower); normal-only gave 0.532. This suggests the differential scheme can be approximated by one-sided acquisition of the higher-signal inverted target, a route to shorter exposure times.