Improved entropy coding for quantized transform coefficients in HEVC screen content coding

In the emerging high efficiency video coding (HEVC) standard, a Golomb-Rice code is adopted to binarize level information. The Golomb-Rice code is optimal for encoding symbols with the exponential probability distribution. In general, quantized transform coefficients of natural images show the exponential probability distribution. However, screen contents usually have a totally different probability distribution, and the Golomb-Rice code is not appropriate for screen content coding. In this paper, we focus on a new entropy coding scheme for screen content coding. In the proposed scheme, we clip inefficient high magnitude coefficient levels that do not fit the exponential probability distribution. Then, we apply a limited length Golomb-Rice code to binarize clipped levels. From experimental results, it is verified that the proposed method achieves on average 0.60% BDrate saving and up to 1.13% BD-rate saving, compared to HEVC screen content coding. When the proposed method is combined with a well-known screen content coding technique, transform skipping, it shows up to 24.02% BD-rate saving.