Absil, Pierre-AntoineJacques, LaurentLambein, XavierXavierLambein2025-05-142025-05-142025-05-142018https://hdl.handle.net/2078.2/7368In the past couple of decades, the discovery of planets outside our solar system has become a relatively common event. Most of them are found by indirect detection, mainly with the radial velocity and transit methods. More recently, a few extrasolar planets have been detected through direct means, by observing light produced by or reflected on the planets themselves. Unfortunately, direct observations are plagued with atmospheric and instrumental noise that decreases the maximum angular resolution we can achieve. Most of the recent advances in this field aim to find ways to eliminate this so-called ``speckle noise''. Angular differential imaging (ADI) offers a solution in which multiple observations of the same system are interspersed with a rotation of the field of view (FOV). The quasi-static speckle noise stays relatively constant between two observations, while the potential planets move through the FOV. With adequate post-processing, ADI can dramatically increase the detectability of planets. We introduce a new ADI post-processing technique in which data is removed around the path of a postulated planet, before being imputed back with a matrix completion method. This model is then compared to the original observations to assess whether the hypothesized planet exists. This new technique is studied with various matrix completion methods, among them a graph-regularized matrix factorization. While the results are good, they fail to improve on the state of the art. However, the idea of using a graph regularizer in low rank models without completion seems promising.Matrix completionExoplanetDirect detectionHigh contrast imagingAngular differential imagingtext::thesis::master thesisthesis:14665