Digital holography made it possible to capture and reconstruct holograms in a computer environment. In a conventional CPU, the real-time reconstruction is not possible when the size of the holograms increase to several mega-pixels range. However, a graphics processor can provide the required computational power. The rapid developments in commercial graphics card technology provide an opportunity to process large blocks of data in a very short amount of time which reduces the hologram reconstruction time significantly. In this manuscript, basics of GPU programming for hologram reconstruction is introduced, and the efficiency of CPU and GPU implementations of the three reconstruction algorithms (Fresnel transformation, angular spectrum method, convolution with free space propagation) are compared. Experimental results indicate that, on average, 100 fps reconstruction rate is achieved with all methods.
Digital holography allows the acquisition of 3D profiles of objects. Digitally captured holograms are reconstructed
at the respective distances of the objects to reveal the phase and the intensity profiles. However, computing an
object’s 3D profile with only a single reconstruction requires prior knowledge of the distance of the object from
the camera. Otherwise, by performing several reconstructions at different distances and by evaluating each image
with sharpness estimation, one can determine the in-focus distance of the object. Moreover, it is not practical
to perform several reconstructions in real-time systems since reconstruction is the most computationally heavy
part in digital holographic imaging. In this paper, we compare common sharpness functions applied to digitally
recorded holograms for autofocus algorithms found in the literature. In addition, we show that automatic focus
distance search can be done in real-time with scaled-down holograms obtained from the original hologram. This
new method improves the speed of autofocus algorithms on the order of square of the scaling ratio. We show
that numerical simulations and experimental results are in good agreement.