Convolutional neural network (CNN) has great potentials in holographic reconstruction. Although excellent results can be achieved by using this technique, the number of training and label data must be the same and strict paired relationship is required. Here, we present a new end-to-end learning-based framework to reconstruct noise-free images in absence of any paired training data and prior knowledge of object real distribution. The algorithm uses the cycle consistency loss and generative adversarial network to implement unpaired training method. It is demonstrated by the experiments that high accuracy reconstruction images can be obtained by using unpaired training and label data. Moreover, the unpaired feature of the algorithm makes the system robust to displacement aberration and defocusing effect.

We propose a method for generating an elliptic perfect vector beam (EPVB) by modulating the dynamic and geometric phases. It is theoretically demonstrated that the shape of the beam can be changed from circle to ellipse by setting the scale factor m of the dynamic phase, but the diameter of it is independent on the topological charge and the polarization order. Since the geometric phases provided by the dialectic Q-plate vary with the polarization state of the illumination beam, EPVB can be converted to the elliptic perfect optical vortex (EPOV) beam by changing the polarization state of the illuminating beam. Therefore, we also provide an alternative method to generate the EPOV beam. The experimental results agree well with the theoretical expectations.

We propose a method to generate arbitrary perfect Poincaré beams using one spatial light modulator (SLM). A polarization beam splitter is used to generate two polarization beams that are orthogonally (x and y) polarized. Half of the SLM helps shape the x-polarized beam; simultaneously, the other half of the SLM shapes the y-polarized beam. By adding different or the same spiral phases to the orthogonally polarized beams, any point on the Poincaré surface can be reached. The proposed approach is experimentally evaluated. The ring diameter remains the same when different polarization orders and topological charges are programmed. The generation of perfect Poincaré beams including fundamental Poincaré sphere, high-order Poincaré sphere, and hybrid Poincaré sphere are demonstrated in our system.

In this paper, we present a phase-shifting-free method to improve the resolution of digital holographic microscopy (DHM) under the structured illumination (SI). The SI used in the system is different from the traditional SI for it is free of the visible structure due to two illumination lights with orthogonal polarization states. To separate the recorded information and also retrieve the object phase, two reference beams with different carrier frequencies and orthogonal polarization states are adopted. The principle component analysis (PCA) algorithm is introduced in the reconstruction process. It is found that the modulated frequency of SI besides the quadratic phases of the imaging system can be easily removed with help of PCA. Therefore, phase-shifting is not required both in recording and reconstruction process. The simulation is performed to validate our method, while the proposed method is applied to the resolution enhancement for amplitude-contrast and phase-contrast objects imaging in experiments. The resolution is doubled in the simulation, and it shows 78% resolution improvement in the experiments.