Optimizing speed and accuracy of sub-pixel phase correlation methods for remote sensing

Correlation of optical satellite images has proved a powerful method for constraining ground deformation associated with natural hazards, such as earthquakes, volcanic eruptions, and landslides. Different approaches to image correlation exist. Spatial correlation methods typically involve multiple evaluations of the correlation coefficient between a reference and a moving image window in space. The resulting correlelogram is then evaluated over a finer (sub-pixel) grid sampling to identify the peak correlation coefficient and thus the best pixel shift in row/EW and column/NS directions. While this method has been shown to be very powerful at resolving ground displacements from optical satellite imagery with high accuracy, precision, and spatial detail, it comes at the expense of computational cost. Furthermore, it generally relies on spatial regularization methods to minimize correlation noise, which can also limit the level of spatial detail we can resolve. Alternatively, phase correlation methods typically involve solving for a pixel shift in the frequency domain (from the phase difference between the two images), and can be significantly faster than spatial correlation methods since it does not rely on a convolutional process. Nevertheless, phase correlation is also sensitive to high frequency noise, which can degrade the estimation of the pixel shift.

In this project, we will explore different methods (including data driven approaches) to estimate the sub-pixel shift in the frequency domain, especially in the presence of noise. The project will focus on optimizing the phase correlation approach for speed (e.g. for GPU processing), as well as accuracy. The main goal of the project is thus to develop a fast and robust open-source phase correlation approach for the remote sensing community, which is capable of processing large volumes of satellite data for the monitoring of Earth surface dynamics. Where appropriate, additional experimentation will be made to refine the correlation approach, e.g. exploring different approaches to phase unwrapping, using multiple correlation iterations, employing spatial regularization techniques, and jointly solving for pixel shifts across multi-band images.

We are looking for a student with a keen interest in remote sensing and natural hazards, and who enjoys programming and code optimization. A background in image processing, geomatic engineering, applied maths or physics/geophysics, would be desirable. The student will work with different researchers at ISTerre, with a rich experience in satellite geodesy, and application of these techniques to Earth Science problems.

For more information, please contact either James Hollingsworth (james.hollingsworth univ-grenoble-alpes.fr) or Sophie Giffard-Roisin (sophie.giffard univ-grenoble-alpes.fr).

Key reference:
Tong, X., Ye, Z., Xu, Y., Gao, S., Xie, H., Du, Q., ... & Stilla, U. (2019). Image registration with Fourier-based image correlation: A comprehensive review of developments and applications. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 12(10), 4062-4081.