Bernhard Schuur

Frequency Selective Update for Video Coding

In this paper the so-called frequency selective update (FSU) technique is presented. FSU is based on a hybrid video coding scheme, where the prediction error is coded and used to correct the prediction signal. This correction of the prediction signal is called update. Standardized video coding schemes such as H.264/AVC apply a transformation, quantization and entropy coding to code the prediction error. In order to reduce the bitrate while keeping the same subjective picture quality, so-called quantization matrices are applied to adapt the quantization to the human visual system (HVS). FSU further enhances the current exploitation of the HVS by selectively updating only few frequency parts. Furthermore, the entropy-coding scheme is adapted to the selected frequency parts and the selection is alternated on frame-by-frame basis. Compared to H.264/AVC high profile, which includes quantization matrices, bitrate reductions of 12-25% are obtained at very similar subjective picture quality.

Advances in hybrid video coding

Hybrid video coding is known for a long time and is applied in all video coding standards such as MPEG-x or H.26x. This paper shows that there is still enough potential for further coding efficiency improvements. The paper starts with an overview of state-of-the-art hybrid video coding schemes such as H.264/AVC. Thereafter, our advances on main building blocks of H.264/AVC are presented that significantly improve the coding efficiency. For instance, intra prediction is improved by changing scan directions and thus providing better reference pixels for specific prediction directions. Adaptive filtering and high precision motion compensation improves the motion compensated prediction. Furthermore, the combination of transformation, quantization, and entropy coding of the prediction error is improved using an advanced frequency selective coding technique. With the transmission of post-filter hints it is possible to design a Wiener post-filter that significantly enhances the picture quality. Finally, texture synthesis techniques are used to improve the subjective quality for specific textures with homogenous statistical characteristics. This paper presents our above-mentioned techniques in detail. Depending on the input sequence and the bit rate, the objective and/or subjective gains compared to H.264/AVC are quite significant.