Abstract: Recent astronomical observations open the possibility to directly image circumstellar disks, a key feature for our understanding of extra-solar systems. However, the faint intensity of these celestial signals compared to the brightness of their hosting star make their accurate characterization a challenging processing task. While there exist several post processing techniques adapted to the extraction of (point shaped) exoplanets, this work proposes a novel iterative method for extracting the elongated structure of circumstellar signals from direct imaging datasets. This method leverages low-complexity models of both the corrupting stellar light and the disks. Our approach is assessed through numerical experiments.