Complexity Management of H.264/AVC Video Compression

The H. 264/AVC video coding standard offers significantly improved compression efficiency and flexibility compared to previous standards. However, the high computational complexity of H. 264/AVC is a problem for codecs running on low-power hand held devices and general purpose computers. This thesis presents new techniques to reduce, control and manage the computational complexity of an H. 264/AVC codec. A new complexity reduction algorithm for H. 264/AVC is developed. This algorithm predicts "skipped" macroblocks prior to motion estimation by estimating a Lagrange ratedistortion cost function. Complexity savings are achieved by not processing the macroblocks that are predicted as "skipped". The Lagrange multiplier is adaptively modelled as a function of the quantisation parameter and video sequence statistics. Simulation results show that this algorithm achieves significant complexity savings with a negligible loss in rate-distortion performance. The complexity reduction algorithm is further developed to achieve complexity-scalable control of the encoding process. The Lagrangian cost estimation is extended to incorporate computational complexity. A target level of complexity is maintained by using a feedback algorithm to update the Lagrange multiplier associated with complexity. Results indicate that scalable complexity control of the encoding process can be achieved whilst maintaining near optimal complexity-rate-distortion performance. A complexity management framework is proposed for maximising the perceptual quality of coded video in a real-time processing-power constrained environment. A real-time frame-level control algorithm and a per-frame complexity control algorithm are combined in order to manage the encoding process such that a high frame rate is maintained without significantly losing frame quality. Subjective evaluations show that the managed complexity approach results in higher perceptual quality compared to a reference encoder that drops frames in computationally constrained situations. These novel algorithms are likely to be useful in implementing real-time H. 264/AVC standard encoders in computationally constrained environments such as low-power mobile devices and general purpose computers.