Adaptive interpolation filters and high-resolution displacements for video coding

Standardized hybrid video coding systems are based on motion compensated prediction with fractional-pel displacement vector resolution. In the recent video coding standard H.264/AVC, a displacement vector resolution of 1/4-pel is applied. In order to estimate and compensate these fractional-pel displacements, interpolation filters are used. So far, these interpolation filters are invariant. The same filter with the same filter-coefficients is applied for all sequences and for all images of a sequence. Therefore, it is not possible to consider nonstationary statistical properties of the video signal and the motion compensated prediction process like aliasing, quantization errors, and displacement estimation errors. This paper presents two techniques that improve the motion compensated prediction and accordingly the coding efficiency. The first technique applies an adaptive interpolation filter that uses filter-coefficients which are adapted once per image. The adapted filter-coefficients are coded and transmitted. The second technique is based on an increased displacement vector resolution of 1/8-pel. In combination, the developed techniques indicate bit-rate reductions up to 25 % compared to H.264/AVC.

I. Introduction

Standardized hybrid video coding systems like the recent H.264/AVC coding scheme [1]–[3] as well as the older standards like MPEG-1 [4], H.262/MPEG-2 [5], H.263 [6], and MPEG-4 Part 2 Visual [7] are based on motion compensated prediction [8], [9]. Fig. 1 shows a generalized block diagram of such a hybrid video encoder.