Low-Complexity MPEG-4 Video Coding Techniques Using Spatial/Temporal Correlation

Abstract

One of the common goals in designing many media coding standards such as MPEG-1, MPEG-2, MPEG-4 Part 2 Visual, and MPEG-4 Part 10 AVC is to increase coding efficiency. For further improving coding efficiency, these coding standards have exploited new coding tools (e.g., rate-distortion optimization, quarter-pixel motion compensation, context-based adaptive binary arithmetic coding (CABAC) in the AVC/H.264 standard) but at the expense of increased encoding complexity. Therefore, the reduction of this complexity is an essential task for designing a fast encoder and decoder. In this dissertation, improved encoding and/or decoding methods are proposed considering computational complexity.

The latest video coding standard, MPEG-4 Part 10 AVC/H.264, provides the best performance in compression efficiency but at the cost of high computational complexity in selecting the best mode during the mode decision process. To reduce this complexity, two fast mode decision algorithms based on spatio-temporal correlation are proposed. The first algorithm exploits the spatio-temporal correlation of macro block (MB) mode types. The second decision algorithm includes the evaluation of the rate-distortion costs of the spatio-temporally neighboring MBs and the current MB. The proposed methods match the best mode decision process described in the AVC/H.264 joint model (JM) 11.0 in compression efficiency but achieve substantially lower computational complexity. The experimental results showed that the AVC/H.264 encoder using the first proposed method produced about 56 % speedup with on average 1.69 % of bit-rate increase and 0.07 dB loss of PSNR compared to the AVC/H.264 JM 11.0 encoder. The AVC/H.264 encoder using the second method achieved a 73 % speedup while preserving a 0.25 % increment in total encoding bit-rates with a loss of only 0.09 dB in PSNR compared to the AVC/H.264 JM 11.0 encoder.

The MPEG committee started to support high resolution video (HRV) by extending the existing standards. Due to the high resolution of HRV, the amount of video data is large and fast encoding and decoding processes are mandatory. In particular, the bigger the video resolution, the more costly the real-time decoding process. In order to reduce the computational complexity of the decoder, a fast entropy decoding method is also proposed in this dissertation. The proposed method was implemented on an MPEG-4 simple profile (SP) codec. The experimental results showed that the proposed method produced about 31 % lower total decoding time and about 71 % (up to 85.9 %) lower entropy decoding time while maintaining similar coding efficiency compared to the MPEG-4 SP codec.