Elham Sakhaee, Ph.D. student, and Alireza Entezari, an associate professor at the CISE, won the best paper award, sponsored by the Mitsubishi Electric Research Lab (MERL) for their paper entitled “Learning Splines for Sparse Tomographic Reconstruction” that was presented at the 10th International Symposium on Visual Computing (ISVC’14).
Tomography is a fundamental computational approach to a wide range of inverse problems that arise in medical sciences, biology, quantum computing, astrophysics and materials science. While classical solutions are effective when there are ample data, they often fail in many practical scenarios, where the data is partially available or incomplete. Our recent research has resulted in a number of contributions to the emerging area of compressed sensing in tomography that aims to enable recovery from partial measurements.
The novel contribution of this paper is to employ a machine learning technique for tomographic image recovery, where the images are represented in spline basis as opposed to commonly used pixel representation. Sparse spline representations are learned from a (noisy) initial reconstruction from limited measurements and enhanced through the proposed iterations consisting of learning and optimization. Our spline-based tomographic system leverages the approximation power of higher order methods and provides significantly higher recovery rates from limited data.
For the interested readers, the paper can be found online in Advances in Visual Computing, Lecture Notes in Computer Science, Volume 8887, 2014, pp 1-10, here.
In addition to spline-based image recovery, we have achieved significant improvements on image restoration from incomplete measurements, through a novel formulation in gradient domain. The interested readers can find the abstracts here and here and access the full papers in proceedings of ICASSP 2015 and ISBI 2015 conferences.