Improved Probability Modelling for Exception Handling in Lossless Screen Content Coding

Competitive methods for lossless screen content coding are based on modelling of probability distributions. The most effective approach for losslessly compressing images with up to 90 000 colours is known as `soft context formation' (SCF). It scans the image to be compressed for repeating patterns and creates corresponding colour histograms. This yields excellent compression results as long as the actual pattern-based histogram contains the colour of the next pixel to be processed. If this colour is not contained, an exception handling is required by sending a special non-colour symbol (escape).This paper proposes an enhanced version of this exception handling coding. Instead of only using a pattern-based estimation, the probability of the escape symbol is now additionally modelled based on local dependencies of exception events. The combination of both estimates leads to 2.0% decrease of bitrate. In comparison to FLIF, FP8v3, and HEVC (HM-16.20+SCM-8.8), the entire SCF method achieves savings of about 26%, 8%, and 12% on average for images with less than 90 000 different colours (for 720p format).