Interferenceless coded aperture correlation holography based on Deep-learning reconstruction of Single-shot object hologram

In interferenceless coded aperture correlation holography with incoherent illumination(I-COACH), point spread hologram(PSH) is important and it is usually necessary to record the PSH library priori. However, the recording of PSH library is time-consuming and basiclly difficult to obtain ideal PSH. The reconstructions correspondingly suffer from some noise which results from the cross-correlation reconstruction of nonideal PSH and object ho-logram (OH). Here a deep-learning-based interferenceless coded aperture correlation holographic imaging technique (DP-based I-COACH) is developed, in which the object can be reconstructed directly from a single-shot object hologram (OH) without any point spread hologram priori. In DP-based I-COACH, a convolutional neural network (CNN) composed of five encoders and four decoders which follows the encoder-decoder "U-net" ar-chitecture is employed. Different object intensity patterns recorded by single-shot together with their associated ground truth form data pairs, are used to train the CNN. In order to demonstrate the reliability of our proposed method, the imaging performances of our proposal is investigated under different experimental conditions, the reconstruction image quality is obviously improved compared with other reconstruction algorithms. The depth of field extension of our proposal without sacrificing the imaging quality and increasing the complexity of system is also described, which will drive the application of I-COACH in some potential scenarios, such as endoscopic application.