“Computational Imaging for VLBI Image Reconstruction,” by Katherine L. Bouman, Michael Johnson, Daniel Zoran, Vincent Fish, Sheperd Doeleman and William Freeman, was submitted in December 2015 and presented at the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). It introduced an algorithm the authors named CHIRP - Continuous High-resolution Image Reconstruction using Patch priors - aimed at the problem of imaging a black hole with the Event Horizon Telescope.
Very long baseline interferometry (VLBI) combines radio telescopes spread across the Earth to act as one giant instrument, but the data it produces is extremely sparse and noisy, and many different images can fit the same measurements. CHIRP framed reconstruction as a Bayesian computational-imaging problem, borrowing statistical “patch prior” image models from the computer-vision community to favor the kinds of structure that look like real images while staying faithful to the data. It is a statistical inference method rather than a deep neural network, which matters when describing how the black hole image was actually made.
The paper drew on ideas from machine learning and computer vision and became part of the toolkit and validation thinking behind the first black hole image, released by the EHT collaboration in April 2019. CHIRP and Bouman’s role in the imaging effort made the project a widely cited bridge between computer-vision research and observational astrophysics.