Complexity-Constrained Video Bitstream Shaping

In scalable or layered video coding, bitstreams that fulfil specific rate or distortion profiles can be derived (shaped) postencoding, i.e., at transmission time. This can also be a very effective method for real-time adjustment of the decoding complexity to bounds specified by the decoding processor or operating system. In this paper, a new type of bitstream shaping is discussed, which is able to fulfil various instantaneous complexity profiles/constraints imposed by the decoder. Complexity is estimated by modeling generic metrics that can be translated into receiver-specific execution-time or energy-consumption estimates. For various bitstream-shaping alternatives, we associate their corresponding complexity estimates with the decoder modules via a decomposition into complexity coefficients and complexity functions. This process drives bitstream shaping according to joint distortion and complexity constraints, while the proposed model incurs limited online computational and storage overhead. In addition, our experiments demonstrate that the proposed method that is based on offline training can outperform various online training methods commonly used for resource prediction in the literature. For each group of pictures, our bitstream-shaping experiments indicate that the accuracy of the proposed method permits adaptation to real-time imposed constraints on the complexity metrics within a 10% error margin