Skipping the transform for impulse-like noise superposed to correlated prediction errors to improve the coding of screen content

For the coding of screen content, Versatile Video Coding, VVC, uses hybrid coding applying prediction and sub-sequent block-wise coding of the prediction error. Blocks with highly correlated prediction errors are transform coded. The transform reduces the correlation and thus the data rate. In the case of screen content, impulse-like prediction errors often occur, which show only marginal or no correlation. Blocks with this impulse-like noise are coded directly without a transform as the transform would generate statistical dependencies between the coefficients and thus would lead to increased data rate. As the content to be coded is typically not aligned with the block structure of the prediction error coding, also situations occur, in which the correlated prediction errors are superposed by impulses. In these situations, none of the two codings is efficient. Transform coding is inefficient for the impulse-like noise and the direct coding is inefficient for the correlated prediction errors. In order to overcome this drawback in this paper, the superposed impulse-like noise is first separated from the correlated prediction errors. After separation, the impulse-like noise is efficiently coded directly without a transform and the correlated prediction errors are coded efficiently by the use of the transform. Experiments using the VVC reference implementation VTM-9.3 show average data rate reductions at the same PSNR of up to 1.6% depending on the sequence.