Posters
« Back
Neural network based Inversion of the quantitative phase inversion problem
EP26422
Neural network based Inversion of the quantitative phase inversion problem
Submitted on 13 Sep 2017

Ayan Sinha, Ons M' Saad, Justin Lee, George Barbastathis
MIT
This poster was presented at Focus on Microscopy
Poster Views: 30
View poster »
Poster Abstract
The phase retrieval problem in optics seeks to recover the phase of a coherent light field given time-averaged intensity measurements. We demonstrate a deep learning technique for neural networks to “learn” solutions to the problem of phase retrieval from an intensity focal stack. We approach the phase retrieval inverse problem from the perspective of model-free sensing: instead of trying to linearize the inverse problem or derive system equations, we let the network attempt to learn in a data-driven manner what those underlying (nonlinear) equations and relationships may be. Specifically, we train our neural networks on a generalized database of natural images and demonstrate in simulation that our technique performs competitively against other focal-stack phase retrieval techniques at varying levels of noise.

[1] L. Waller, L. Tian, G. Barbastathis, Optics Express 18(12), 12552-12561 (2010).
[2] Z. Jingshan, R. Claus, J Dauwels, L. Tian, L. Waller, Optics Express 22(9), 10661-10674 (2014).
[3] K. He, X. Zhang, S. Ren, and J. Sun, IEEE Conference on CVPR, (2016).
Report abuse »
Questions
Ask the author a question about this poster.
Ask a Question »

Creative Commons