Context-based adaptive variable length coding for video dct blocks part ii - practical scheme

This paper (for Part I, see [8]) proposes a new context-based adaptive variable length coding scheme (CBAVLC) for video DCT blocks, aiming to maximize the achievable coding efficiency under acceptable complexity in practical video coding. CBAVLC is designed by following the theoretically optimal approach described in Part I. The proposed context modeling in CBAVLC is domain knowledge drove, which fully explores the observed domain properties of DCT blocks for high efficiency. The adaptive coding in CBAVLC uses Golomb-Rice codes and Elementary Golomb codes. The adaptation rules come from the famous LOCO-I algorithm, which are approximations of theoretically optimal ones. The experimental results show that on the test videos CBAVLC significantly improves coding efficiency relative to 2D-VLC, and compared to CAVLC a few coding efficiency gains are also obtained.