Real-time, auto-focusing digital holographic microscope using graphics processors

Official URL: http://dx.doi.org/10.1063/1.4818285

The most significant advantage of holographic imaging is that one does not need to do focusing alignment for the scene or objects while capturing their images. To focus on a particular object recorded in a digital hologram, a post-processing on the recorded image must be performed. This post-processing, so called the reconstruction, is essentially the calculation of wave propagation in free space. If the object's optical distance to the recording plane is not known a priori, focusing methods are used to estimate this distance. However, these operations can be quite time consuming as the hologram sizes increase. When there is a time constraint on these procedures and the image resolution is high, traditional central processing units (CPUs) can no longer satisfy the desired reconstruction speeds. Then, especially for real-time operations, additional hardware accelerators are required for reconstructing high resolution holograms. To this extend, today's commercial graphic cards offer a viable solution, as the holograms can be reconstructed tens of times faster with a graphics processing unit than with the state-of-the-art CPUs. Here we present an auto-focusing megapixel-resolution digital holographic microscope (DHM) that uses a graphics processing unit (GPU) as the calculation engine. The computational power of the GPU allows the DHM to work in real-time such that the reconstruction distance is estimated unsupervised, and the post-processing of the holograms are made completely transparent to the user. We compare DHM with GPU and CPU and present experimental results showing a maximum of 70 focused reconstructions per second (frps) with 1024 x 1024 pixel holograms.