Hide/Show Apps

Development of novel analysis and reconstruction techniques for coherent optical imaging systems

Işıl, Çağatay
We develop novel analysis and reconstruction techniques for coherent optical imaging systems. Firstly, we present a phase-space approach to analyze coherent imaging systems with multiple diffracting apertures. The degrees of freedom of a coherent imaging system can be computed from its phase-space window, which takes into account diffraction effects from all apertures. We show how the phase-space window is linked to important imaging parameters of the system such as diffraction-limited resolution. A single-lens system and a microscope objective design are considered as examples to illustrate the utility of the approach. Secondly, we focus on the classical phase retrieval problem, which is a fundamental problem in coherent imaging. Although there are several well-known phase retrieval algorithms, the reconstruction performance is generally sensitive to initialization and measurement noise. We develop two different novel phase retrieval algorithms by jointly exploiting deep neural networks (DNNs) and traditional model-based inversion methods. The used model-based inversion approach is the well-known hybrid-input-output (HIO) method for phase retrieval. In the first approach, the main idea is to use a DNN in an iterative manner with the HIO method to improve the HIO reconstructions. Numerical results demonstrate the effectiveness of this approach, which also has little additional computational cost compared to the HIO method. In the second approach, the main idea is to incorporate a learning-based prior to the HIO method through plug-and-play regularization. The developed method is flexible such that it can also be used with different image priors. The performance of the second approach is illustrated with numerical simulations. Both of the developed phase retrieval methods not only achieve state-of-the-art reconstruction performance but also are more robust to different initialization and noise levels.