In this work, we use the High-Dimensional Mixture model previously introduced for image denoising (HDMI) to solve inverse problems of increased complexity. Since the HDMI model uses dimension reduction regularization, we are able to learn it even in the case of missing data. Thus, we can extend the model to the case of missing pixel restoration. We also explore the use of the HDMI model for image reconstruction problems such as inpainting.

This is poster number 50 in Poster Session

Authors:

Antoine Houdard (Télécom ParisTech)

Julie Delon (Université Paris Descartes)

Keywords:

image reconstruction, inverse problems