• home
• current students
• graduate
• events

Presenter: Mehrnaz Movahed
Supervisor:

Date: Thu, January 16, 2020
Time: 16:30:00 - 17:30:00
Place: EOW 230

ABSTRACT

Abstract:

Synthetic aperture radar (SAR) data has been an interesting subject to study and investigate for researchers. One of the applications of SAR data is obtaining interferometry synthetic aperture radar (InSAR) images. However, these images as they are, are not useful for research purposes. Similar to other applications involving images, the presence of noise can distort a significant amount or all of the desired information. Various processes are applied to these images to extract the desired information. One of the pieces of information that is extracted is phase image. Because of the periodic nature of phase, phase images only contain values between (-π,π]. Thus, resulting in an ambiguous phase image. Phase unwrapping is applied to recover the true value of the phase from the phase image. Many algorithms have been proposed for phase unwrapping. Some of these algorithms, such as belief propagation (BP) and graph cuts, are energy minimization algorithms. That is, they convert the phase unwrapping into energy minimization and find a lower bound on the energy function. In the process, these algorithms find the value of the real phase in the phase images. However, these algorithms suffer from various drawbacks such as limited number of applicable energy functions, getting stuck in a loop, or relatively low accuracy. To overcome some of these drawbacks, sequential tree re-weighted algorithm (TRWS) was proposed. TRWS has better but still relatively low accuracy. In this seminar, we present how the accuracy of TRWS can be improved. To achieve better accuracy we use more dense graphs (graphs with more edges) than what was originally proposed in TRWS. To compare the performance of our proposal with TRWS in terms of accuracy we tested our algorithm on synthetic images similar to actual InSAR images. Simulation results show that our proposal has more accuracy than TRWS. In particular, the mean square error value (obtained from comparing the simulation results with the true value) can be improved by more than 98%. This shows that our proposal has a much better performance than TRWS for phase unwrapping.