Dynamic Fourier ptychography with deep spatiotemporal priors

Abstract Fourier ptychography (FP) involves the acquisition of several low-resolution intensity images a sample under varying illumination angles. They are then combined into high-resolution complex-valued image by solving phase-retrieval problem. The objective in dynamic FP is to obtain sequence moving sample. There, application standard frame-by-frame reconstruction methods limits temporal resolution due large number measurements that must be acquired for each frame. In this work instead, we propose neural-network-based framework FP. Specifically, reconstructed output shared deep convolutional network fed with an input vector lies on one-dimensional manifold encodes time. We optimize parameters fit measurements. architecture and constraints vectors impose spatiotemporal regularization images. This enables our method achieve high without compromising spatial resolution. proposed does not require training data. It also recovers pupil function microscope. Through numerical experiments, show paves way high-quality ultrafast