A video decoder is configured to reconstruct a first version of a current picture; determine a transformed sample of the first component, a transformed sample of the second component, and a transformed sample of the third component by applying a first matrix to a triplet comprising the samples of the first, second, and third components; include the transformed samples of the first, second, and third components in a second version of the current picture; determine, based on the transformed samples of the first, second, and third components and a first row of a second matrix, a set of transformed samples for the third component, the set of transformed samples for the third component including samples of the third component that are immediately neighboring the transformed sample of the third component; and include the set of transformed samples for the third component in the second version of the current picture.