Cleaning Images with Gaussian Process Regression

Many approaches to astronomical data reduction and analysis cannot tolerate missing data: corrupted pixels must first have their values imputed. This paper presents astrofix, a robust flexible image imputation algorithm based on Gaussian process regression. Through an optimization process, astrofix chooses applies different interpolation kernel each image, using training set extracted automatically from that image. It naturally handles clusters of bad edges adapts various instruments types. For bright pixels, the mean absolute error is several times smaller than median replacement by kernel. We demonstrate good performance both imaging spectroscopic data, including SBIG 6303 0.4 m telescope FLOYDS spectrograph Las Cumbres Observatory CHARIS integral-field Subaru Telescope.