Real-time reference-based dynamic phase retrieval algorithm for optical measurement.

Abstract

To study dynamic behaviors of a phenomenon, measuring the evolving field of a specimen/material/structure is required. Optical interferometry, as a full-field, non-contact, and highly sensitive optical measurement technique, has been applied, where the evolving field is represented as dynamic phase distribution. A dynamic phase retrieval algorithm, called least-squares with 3 unknowns (LS3U), which estimates the phase change between each two consecutive patterns by a least-squares fitting method and denoises the phase change by a windowed Fourier filtering (WFF) algorithm, has been shown to be a simple yet effective algorithm. However, LS3U is computationally expensive, restricting its potential application in real-time dynamic phase retrieval systems. In this paper, a real-time LS3U algorithm powered by GPU parallel computing is proposed, with which frame rates of up to 64.5 frames per second (fps) and 131.8fps are achieved on NVIDIA's GTX 680 and GTX 1080 graphics cards, respectively.